EXPENSE NEGLECT
1
Expense Neglect in Forecasting Personal Finances
Jonathan Z. Berman
An T. K. Tran
John G. Lynch, Jr.
Gal Zauberman
AUTHORS’ NOTE
Jonathan Z. Berman (jberman@london.edu) is an Assistant Professor of Marketing at the
London Business School; An T. K. Tran (att57@drexel.edu) is an Assistant Clinical Professor of
Marketing at the LeBow College of Business, Drexel University; John G. Lynch, Jr.
(john.g.lynch@colorado.edu) is the Ted Andersen Professor of Free Enterprise at the Leeds
School of Business, University of Colorado Boulder; and Gal Zauberman
(gal@wharton.upenn.edu) is the Laura and John J. Pomerantz Professor of Marketing and
Professor of Psychology at The Wharton School, University of Pennsylvania. The authors would
like to thank members of the Consumer Financial Decision Making Lab group at the University
of Colorado and seminar participants at Erasmus University, INSEAD, UC-Berkeley, Yale
University, the Utah Winter JDM Symposium, and the University of Pennsylvania.
Correspondence concerning this article should be addressed to the first author.
EXPENSE NEGLECT
2
ABSTRACT
This paper shows evidence for expense neglect in how consumers forecast their future spare
money or “financial slack.” Even though people generally think that both their income and
expenses will rise in the future, we find that they systematically under-weigh the extent to which
their expected growing expenses will cut into their spare money (Studies 1-9). We rule out the
possibility that these findings are due to: measurement error (Studies 2-5); a lack of confidence
in estimating future expenses (Study 6); belief in greater flexibility of future expenses (Study 7);
or a general optimism bias (Study 8). Study 9 shows that consumers who are chronically attuned
to expenses (tightwads) are less likely to show expense neglect than those who are not
(spendthrifts). Finally, we conduct a meta-analysis of our entire file-drawer (25 studies, 7,214
participants) and find that across all studies participants place about 2.7 times the weight on
income change as they do on expense change when forecasting their financial slack, and that
expense neglect is stronger for the distant than near future.
Keywords: Forecasting, Financial Slack, Personal Finance, Expense Neglect
EXPENSE NEGLECT
3
Successfully managing one’s personal finances is rarely straightforward. It is a complex
problem to balance how much to spend and save each month in order to put one’s hard earned
money to its best use. When making monetary decisions, consumers often have to assess the
overall state of their finances in order to maximize long-term wellbeing. These subjective
assessments depend not just on consumers’ perceptions of their current bank account balances,
but also on how their finances will change over time. Someone who is starting a new job may
wonder whether to put away money for retirement immediately, or if he can afford to put it off
until he gets his next raise. A couple looking to purchase a new home may ask if they will be
able to afford to keep up with mortgage payments in the nicest part of town if and when they
have children. A student with credit card debt must determine whether she will be able to afford
to repay her next purchase plus interest once she graduates and—hopefully—gets a job.
Mispredicting the state of one’s future personal finances can be highly consequential.
Those who mistakenly believe that they will have significantly more spare money in the future
may choose to take out loans they will not be able to repay or may put off saving for retirement
until it is too late (Thaler and Benarzi 2004; Lynch and Zauberman 2006). Those who mistakenly
believe they will have more constrained resources in the future than reality may fail to spend
enough money in the present and will miss out on essential life experiences (cf. Kivetz and
Keinan 2006).
Further, recent research shows that perceived resource availability significantly affects
consumers’ thoughts, preferences, and actions. The more available resources consumers expect
to have in the future, the more likely they discount delayed expenditures of the resource
(Zauberman and Lynch 2005). Those who feel financially constrained are more likely to focus
attention on solving financial problems in the present at the cost of the future (Shah,
EXPENSE NEGLECT
4
Mullainathan, and Shafir 2012) and are more likely to prefer scarce rather than abundant
products (Sharma and Alter 2012).
More broadly, indices such as the GfK NOP Consumer Confidence Barometer, the
Bloomberg-Nanos Canadian Confidence Index, the Conference Board’s Consumer Confidence
Survey, and the University of Michigan Consumer Sentiment Index (MCSI) attempt to measure
consumer’s feelings about the state of their finances in order to provide meaningful inputs for
manufacturers, retailers, and governments to manage their output. Greater consumer confidence
signals that households feel positive about their future finances, and businesses should adjust
their production in order to respond to the likelihood of stronger demand.
These considerations motivated us to study how consumers form subjective assessments
of their finances in the future. Although consumers often feel financially constrained in the
present moment, they often view the future more favorably. We examined the entire MCSI data
set, from 1978 to 2013, and found that Americans consistently anticipated that their finances
would improve in the future. When asked to estimate their family’s finances in a year’s time,
more respondents indicate that they expect their future finances to be better off than worse off in
407 out of 425 months. More recently, in the period from 2010 to 2013—the time period
covering 24 of our 25 studies—more consumers expected that their financial situation to be
better off than worse off in 47 out of 48 months. In contrast, the most recent Federal Reserve
Survey of Consumer Finances shows that during the same period, median net worth of all
families fell by two percent (Bricker et al. 2014). In other words, despite households being on
average worse off in 2013 than they were in 2010, they still consistently forecasted their future
finances favorably over this period.
EXPENSE NEGLECT
5
In the present research, we consider one reason why consumers view their future finances
favorably by examining subjective feelings of financial slack. We define financial slack as the
perceived surplus (or lack) of spare money an individual has at a given point in time (Zauberman
and Lynch 2005). At a fundamental level, the amount of financial slack consumers have at a
given point in time is a function of their income (inflow) and expenses (outflow). Whereas
increasing income contributes to having more financial slack, increasing expenses decreases
financial slack.
The above analysis of the MCSI may suggest that consumers are optimistic about their
future finances, and thereby consistently believe that their finances will improve with time.
However, we explore a different explanation, namely that consumers improperly account for
their expenses (and consequently under-weigh growth in expenses) when forecasting their
finances. Specifically, we examine how consumers utilize their own stated assessments of their
income and expenses to determine how much financial slack they will have in the future.
Our main interest is in evaluating the weight that consumers place on their anticipated
income and expenses in predicting their future financial slack. By weight, we mean the extent to
which consumers consider their own income and expenses expectations to determine how much
financial slack they will have in the future. This question is different from research that examines
how accurate consumers are in their predictions about their income or expense estimates (Peetz
and Buehler 2009, 2012; Sussman and Alter 2012; Ülkümen, Thomas, and Morwitz 2008).
Rather, we’re interested in how consumers form subjective estimates of their future finances,
given their own stated assessment of changes in their future income and expenses.
Objective and Subjective Determinants of Available Financial Resources
EXPENSE NEGLECT
6
Subjective feelings of financial slack represent the extent to which consumers believe that
they are in a surplus or deficit state with respect to their available spare money. When assessing
the future financial slack, income and expenses are equally relevant. It’s simple arithmetic. In
any given period, consumers generate financial resources and spend them. The amount of
available slack that a consumer expects to have in a given period should be related to the stock of
that slack account in the prior period plus inflows of income minus the outflows of expenditures
in the most recent period. Inflows include salary, bonuses, investment returns, social security
checks, interest, etc., whereas outflows include money spent on rent, bills, travel expenses,
entertainment as well as savings commitments, debt payments, donations to charity, etc.
If the mental models of financial slack matched the accounting rules of bank balances, as
it should, then a dollar of inflow would have exactly the same weight in predicting future slack
as a dollar of outflow. For example, a waitress who needs an extra $80 a month in spare money
can achieve her goal either by raising her income (e.g., working an extra shift that will garner her
$80 in after-tax cash) or by cutting her expenses (e.g., suspending her $80/month cable package).
That is, the weights of inflows (income) and expenses (outflows) in affecting slack should be
equal and opposite in sign. However, we find that people systematically under-weigh their own
stated expenses relative to income, and therefore rely too heavily on forecasts of income when
predicting their future finances.
Because income and expenses are weighed equally in actual bank balances, it may not
seem difficult to forecast changes in spare money over time. However, existing research
demonstrates that it is surprisingly difficult to estimate dynamic systems with inflows and
outflows, suggesting that forecasting future spare money may not be so simple. Most people are
flummoxed by even basic “stock-flow” problems, such as predicting the number of people in a
EXPENSE NEGLECT
7
department store with people entering and exiting at different rates (Booth Sweeney and Sterman
2000; Cronin and Gonzalez 2007; Cronin, Gonzalez and Sterman 2009; Sterman and Booth
Sweeney 2002).
Forecasting finances into the future is even a more difficult stock-flow problem than
those studied previously. With financial decisions, expected changes over time in income and
expenses are not independent. Those who get a raise often expect to increase their spending;
those who lose a source of income often expect to cut their spending. Thus, when forecasting
spare money into the future, consumers need to also account for how consumption and expenses
rise with increases in income.
Expense Neglect in Forecasting Financial Slack
In the studies we present, we show that consumers neglect to properly account for their
own stated expenses relative to income when forecasting their financial slack. In each study, we
ask participants to estimate how their income, expenses, and subjective financial slack (i.e.,
“spare money”) will change over time. We then use each individual i's own predictions of how
their stated income and expenses will change over time to predict that person’s own estimates of
how slack will change over time, as represented in equation 1.
(1) Slack Changei = β0 + β1(Income Changei) + β2(Expense Changei) + εi
Under the model in equation 1, three conditions should hold. First, controlling for
expense changes, income changes should be positively associated with slack changes; as income
increases, so does spare money. Second, controlling for income changes, expense changes should
EXPENSE NEGLECT
8
be negatively associated with slack changes; as spending increases, spare money decreases.
Third, and most critically, the weight of a unit change in income change should be equal to the
weight of a unit change in expenses (β1 = -β2), since a dollar increase in income should have the
same impact on financial slack as a dollar decrease in expenses.
However, we find robust evidence that people do not properly weigh their own reported
expense change when estimating their future money slack change. We report in detail 9 of the
studies we have conducted, as well as a meta-analysis of our entire file drawer of 25 studies (N =
7,214). We have four main findings. First, consumers expect their incomes to increase in the
future, and more so in the distant than near future. Second, consumers expect their expenses to
increase in the near future, and expect larger increases in the distant future. Third, consumers
expect to have more financial slack in the near future than today, and expect larger increases in
the distant future.
Our primary finding, though, is that predictions of changes in financial slack are much
more driven by expected changes in income than by expected changes in expense—a pattern we
call expense neglect. When participants are asked to forecast how much financial slack they will
have in the future, participants rely heavily on their own income forecasts and neglect or underweigh their own expense forecasts (i.e., in equation 1, β1 > -β2). In other words, while people do
expect that their expenses will rise slightly in the future, they do not properly adjust their
estimates of their future financial slack to capture these expectations.
Expense Neglect is Neither a Form of the Planning Fallacy nor an Optimism Bias
Before reporting our studies in detail, we next articulate what we are not studying as we
have found in presenting this work that audiences tend to expect our findings to connect to
EXPENSE NEGLECT
9
literatures that are different from our focus. We are not studying some form of a planning or
budget fallacy (Buehler, Griffin and Ross 1994; Kahneman and Tversky 1979; Peetz and Buehler
2009). Overestimating changes in income or underestimating changes in expenses could lead
consumers to falsely expect that spare money will increase in the future. However, our focus in
this research is not on whether consumers are accurate in their predictions of future income,
expenses, or financial slack (cf. Peetz and Buehler 2009; 2012; Sussman and Alter 2012;
Ülkümen, Thomas, and Morwitz 2008). Rather, we are interested in the weighting of income and
expenses to predict financial slack.
Similarly, the effects we show are unrelated to any form of generalized optimism bias (cf.
Puri and Robinson 2007). If consumers are generally optimistic, they may anticipate having
greater income and few expenses than in reality. However, regardless of how accurate consumers
are in their predictions, we can still evaluate how their income and expense estimates feed into
their slack estimates. Expense neglect predicts that consumers will primarily base their estimates
about their future financial slack on their own income estimates, and will not sufficiently account
for their own expense estimates. One might nonetheless argue that expense neglect is a special
case of optimism bias. Perhaps the relative weights of income and expenses in predicting future
spare money are a function of optimism. We examine this directly by measuring individual
differences in optimism in Study 8, and find that while optimistic people expect to have greater
future income and fewer future expenses than pessimists, optimism does not affect the relative
weighting of income or expenses.
OVERVIEW OF STUDIES
EXPENSE NEGLECT
10
Nine studies show that consumers exhibit expense neglect when forecasting their future
financial slack. Study 1 shows evidence for expense neglect in forecasting future slack across a
range of samples taken from participants with varying financial conditions. Studies 2 – 5 show
that these results hold through multiple ways in which slack, income, and expenses are measured.
Studies 6 – 9 examine possible causes of expense neglect. We first show that this effect cannot
be explained by greater confidence in predictions of future income than in predictions of future
expenses (Study 6), beliefs about increasing flexibility regarding future expenses (Study 7), or by
a general optimism bias (Study 8). We test an attention account, and show that those who are
chronically attuned to expenses are more likely to account for expenses in their forecasts of spare
money (Study 9). Finally, we present a meta-analysis of our entire file-drawer, which covers 25
separate studies and 7,214 participants. Across studies we find that consumers place about 2.7
times the weight on standardized income change as they do on standardized expense change, and
that expense neglect is more prominent for forecasting the more distant versus the near future.
STUDY 1: EXPENSE NEGLECT ACROSS VARYING POPULATIONS
Study 1 examines how consumers attend to their expected income change and expense
change when estimating how their financial slack will change with time. We followed Cook and
Campbell’s (1979) research strategy of “deliberate sampling for heterogeneity” by examining
four different groups with varying financial circumstances and different realistic prospects for
improvement in the near future (employed, unemployed, executives, and students).
Method
EXPENSE NEGLECT
11
We conducted three initial surveys in October and November of 2011 across different
populations. Survey 1 was administered to 300 participants (51.7% female; mean age = 47.5) via
a national online panel. This survey sampled equal groups of those who were employed full time,
employed part time, unemployed, and out of the labor force. The first two groups were combined
to create a sample of 150 employed participants (53.3% female; mean age = 46.4) and the latter
two were combined to create a sample of 150 unemployed participants (50.0% female; mean age
= 48.6). Survey 2 was administered to 216 students (52.2% female; mean age = 20.4) at a private
northeastern university. Survey 3 was administered to 64 business executives from an online
panel (15.6% female; mean age = 48.7).
Participants first estimated their spare money over this week, a week in three months, and
a week in twelve months on an 11-point scale ranging from 1 = “Very little spare money” to 11 =
“A lot of spare money”. Next, participants estimated their expected change in income and
expenses in three and twelve months on an 11-point scale anchored by -5 = “I expect my income
[expenses] to greatly decrease”; 0 = “I expect my income [expenses] will remain the same”; and
+5 “I expect my income [expenses] to greatly increase.” In the present study, participants
indicated their expected income change before indicating their expected expense change. In all
subsequent studies, we counterbalance the presentation order of these questions.
Results
Slack, Income and Expense Change: Slack change scores were calculated by subtracting
individual ratings of future slack from ratings of current slack for each time period. Figure 1a
displays expected slack change for each sample, while Figure 1b displays expected changes in
income and expenses for each sample.
EXPENSE NEGLECT
12
---Insert Figures 1a and 1b about here---
On average participants in each sample estimated that their income and expenses will
increase from the present to three months in the future. We reject the null hypothesis that
changes = 0 (all Ms > 0.45; one-sample ts > 3.15, ps < .002). This pattern is even more
pronounced when comparing the present to twelve months in the future. Participants in all
groups expected that their income and expenses to increase more in twelve months as compared
to three months (paired ts > 2.30, ps < .025).
Predictions of Slack Growth: We examined the extent to which participants weighed
income changes versus expense changes when estimating their future financial slack growth.
Across all participants we ran separate regressions for the three months’ and twelve months’ time
periods. Each regression included income change, expense change for the ith respondent in
group j, a dummy code for the separate groups as our independent variables, and slack change as
our dependent variable. We refer to this as the “full model”:
(2) SlackChangeij = β0 + β1(Income Changeij) + β2(Expense Changeij) + β3-5(Groupj) + εij
Results show significant expense neglect. When estimating slack change in the short term
(this week vs. three months), greater income change led to greater slack change (β = .27, SE
= .05; t(574) = 5.41, p < .001), but expected expense change did not significantly influence slack
change (β = .04, SE = .06; t(574) = .69, p = .49). Similarly, when estimating slack change in the
long term (this week vs. twelve months), greater income change lead to greater slack change (β
EXPENSE NEGLECT
13
= .46, SE = .06; t(574) = 8.13, p < .001), but expected expense change did not significantly
influence slack change (β = -.04, SE = .06; t(574) = -.70, p = .49).
We employed the following model comparison test to evaluate whether the coefficient on
income is significantly larger in magnitude than the coefficient on expenses to show that people
under-weigh expenses relative to income. This test compared the fit of the regression above (the
full model) to a “constrained model” that sets income and expenses to have equal and opposite
signs. The constrained model for Study 1 is listed below:
(3) SlackChangeij = β0 + β1(Income Changeij – Expense Changeij) + β2-4(Groupj) + εij
If participants weigh expense change and income change equally, then a constrained
model that requires income and expenses to have equal and opposite signs (equation 3) should
not fit worse than the full model where the parameters on income and expenses are not
constrained (equation 2). However, if participants are weighing income and expenses unequally,
then the full model should fit significantly better than the constrained model.
Results show that the full model that allows the weights of income and expenses to vary
2
independently fits significantly better than the constrained model for both the three month (𝑅𝐹𝑢𝑙𝑙
2
2
= .06 vs. 𝑅𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑒𝑑
= .02; F(1, 574) = 20.74, p < .001) and twelve month (𝑅𝐹𝑢𝑙𝑙
= .12 vs.
2
𝑅𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑒𝑑
= .07; F(1, 574) = 33.48, p < .001) time periods. Clearly, participants are not
attending to income and expenses equally. We further examine the robustness of these results by
evaluating each of the four groups separately. Table 1 displays these results along with the
results from a model comparison that compares the full versus the constrained model for each
group.
EXPENSE NEGLECT
14
---Insert Table 1 about here---
Note that the lower weight of income compared to expenses is not an artifact of
correlation between income and expense predictors. The within-sample correlations between
income growth and expense growth are relatively low (rs < .35), and we see low variance
inflation factors across all regressions (VIFs < 1.14; Cohen, Cohen, West and Aiken 2003). In all
subsequent studies, we find no evidence for multicollinearity affecting our results. Further,
multicollinearlity cannot explain the significantly better fit of equation 2 compared to equation 3.
Discussion
Study 1 demonstrates expense neglect across four samples of markedly different
respondents. Even though participants anticipated their expenses going up in both the near and
distant future, they do not take this into account when forecasting their future spare money. For
the twelve months’ time period, only the executives did not show a significant degree of
expenses neglect, although their pattern of coefficients on income and expenses was similar to
those in the other groups.
STUDIES 2 – 5: MEASURMENT ERROR DOES NOT EXPLAIN EXPENSE NEGLECT
In Studies 2 - 5, we rule out a number of alternative explanations that would suggest
these results are due to ways in which we measure income, expenses, and financial slack.
Specifically, we show that this effect holds when we: (1) explicitly define income and expenses
to each participant; (2) adjust the ‘current’ time period for estimates of income, expense, and
EXPENSE NEGLECT
15
slack from a week to a month which reflects a more natural time frame for evaluating finances;
(3) take estimates of income and expenses at two points in time rather than asking about changes
over time; (4) account for any scale compatibility effects by having participants evaluate
financial slack in terms of “constraint” as opposed to “spare money”; (5) use a multi-scaled
measure of slack that reflects more specified of assessments of financial slack; and (6) estimate
income and expenses using both a modulus and dollar estimates to control for different uses in
the response scale. Each study was run on Mechanical Turk. Additional details pertaining to each
study can be found in Web Appendix A.
Across all subsequent studies in the manuscript, we make the following changes. First,
we counterbalance the order of the slack questions with the income and expense questions.
Second, we explicitly define income and expenses to participants. For the income question,
participants read, “We would like for you to think about your monthly income. This includes all
the money that you will earn from salary, bonuses, social security checks, interest, alimony
receipts, etc.” For the expenses question, participants read, “We would like for you to think
about your monthly expenses. This includes all the money that you spend in a given month such
as your rent, bills, travel expenses, entertainment etc. as well as any savings commitments, debt
payments and donations to charity.” Third, we adjust the time period for estimates of income,
expense, and slack from a week to a month, which reflects a more natural time frame for
participants to evaluate their finances.
Study 2: Estimating Future Financial Constraint Does Not Increase the Weight Placed on
Expenses
EXPENSE NEGLECT
16
Study 2 (N = 302) makes two changes to the design in Study 1, in addition to those
mentioned in the preceding paragraph. We ask participants to make “static” estimates of income
and expenses at two points in time rather than asking participants to report estimate changes over
time. Specifically, participants estimated the amount of income and expenses they expect to have
next month and over a month in two years from now (on 11-point scales anchored by 1 = “Very
little” to 11 = “A lot”).
We also test whether a scale compatibility effect is driving our results (Slovic, Griffin,
and Tversky 1990; Tversky, Sattah, and Slovic 1988). Both income and spare money may be
associated with inflows that create wealth, whereas expenses are associated with outflows that
inhibit wealth accumulation. It may be that when participants think about spare money, they are
more likely to attend to income, an input that contributes to spare money, rather than expenses,
an output that reduces spare money. To test this, half the participants estimated their “financial
constraint”, a measure that is presumably more associated with outputs that limit wealth, whereas
the other half of participants estimated their “spare money”, as in the previous study.
We first examined whether asking participants about their “spare money” versus
“financial constraint” affected the weight of income change and expenses change on slack
change. The financial constraint measure was reverse-coded such that higher numbers indicated
more financial slack. Change scores were computed to determine how much participants
expected their income, expense, and slack to increase or decrease in the future. Next, we ran the
following regression:
(4) Slack Change =
β0 + β1(Response Scale) + β2(Income Changei) + β3(Expense Changei) +
β4(Response Scale*Income Changei) +
EXPENSE NEGLECT
17
β5(Response Scale*Expense Changei) + i
Response scale was contrast coded (-1 = Financial Constraint, +1 = Spare Money). Results show
that neither the response scale by expenses change interaction (t(296) = -.19, p = .85) nor the
response scale by income change interaction (t(296) = .09, p = .93) were significant, suggesting
that the response scale did not affect the weight placed on either income or expense change when
estimating slack. There was also no simple main effect of response scale on slack change when
income change and expense change were coded as zero (t(296) = -.80, p = .42).
We collapsed across response scale and regressed slack change on income change and
expense change to test for expense neglect. We replicate expense neglect, finding that the
coefficient on income change (β = .47, SE = .06, t(299) = 7.98, p < .001) is 2.5 times the size of
the coefficient on expense change (β = -.19, SE = 0.06, t(299) = -3.44, p < .001). Further, the full
model (R2 = .19) that allowed the income change and expense change parameters to vary
explains significantly more variance than the constrained model where income and expenses are
fixed to have equal and opposite signs (R2 = .15; F(1, 299) = 13.60, p < .001).
Study 3: Multi-Item Measures of Financial Slack Does Not Eliminate Expense Neglect
In Study 3 (N = 74), we test for the possibility that participants are interpreting “spare
money” and “financial constraint” more broadly than intended. We account for this possibility by
introducing an additional financial slack measure that asks participants how likely they would be
able to make an emergency repayment of a given amount out of their liquid discretionary funds.
A measure that asks participants about their ability to make a fixed payment out of their liquid
EXPENSE NEGLECT
18
funds is robust to different ways in which people may be thinking about financial slack, and also
excludes any wealth effects that may arise from the purchase of illiquid assets.1
Participants were asked about their personal finances next month and in two years’ time.
We created a three-item measure of financial slack that included the spare money and the
reverse-coded financial constraint measures used in the previous study, and the following
“likelihood of repayment” question: “Imagine that next month [in 24 months] you had an
unexpected expense of $1,500 such as a medical bill or a necessary car repair. How likely is it
that you would be able to pay this bill in full and on time without having to dip into your
retirement fund, borrow money, or charge it to a credit card?” on a scale ranging from 1 = “Very
unlikely” to 11 = “Very likely”.
We replicate expense neglect using a three-item measure of slack change (α = .69). We
first ran a regression with income change and expense change to predict slack change. The
coefficient on income change (β = .81, SE = .13, t(71) = 6.24, p < .001) is 2.0 times the size of
the coefficient on expense change (β = -.41, SE = .13, t(71) = -3.10, p = .003). Further, the full
model (R2 = .36) that allows the income and expense parameters to vary explains significantly
more variance than the constrained model where income and expenses have equal and opposite
signs (R2 = .30; F(1, 71) = 6.88, p = .011).
Importantly, our results replicate if we only evaluate the “likelihood of repayment” item
as the measure of financial slack. In this case, the coefficient on income change (β = .93, SE
= .22, t(71) = 4.23, p < .001) is 3.4 times the size of the coefficient on expense change (β = -.27,
SE = .23, t(71) = -1.19, p = .003), and the full model (R2 = .20) explains a significantly greater
1
For instance, consider an individual who expects to receive a raise in the future, and expects to spend the money on
an illiquid asset, such as a mortgage. It is possible that this person would indicate an increase in income (the raise),
expenses (the mortgage payment) and financial slack (a wealth increase due to the mortgage), a pattern of results
that would appear as if this person was neglecting expenses.
EXPENSE NEGLECT
19
amount of variance than the constrained model (R2 = .13; F(1, 71) = 6.66, p = .012). This
increases our confidence that our core effect is not due to the way that participants were thinking
about the subjective financial slack questions used in the previous studies.
Study 4: Ruling out Response Language Using Magnitude Estimation
In the previous studies, we utilize an 11-point ratings scale to evaluate participants’
estimations regarding their income and expenses. It is possible that consumers are using the scale
to rate their income differently from the way in which they use the scale to rate their expenses
(Lynch, Chakravarti, and Mitra 1991). If so, then we may not be accurately measuring
differential weighting of participants’ true feelings about how their income and expenses will
change in the future.
In Study 4 (N = 244), we examine whether the use of a modulus to measure income and
expense change can account for expense neglect. The use of a modulus is a standard technique in
“magnitude estimation” in psychophysics to keep estimations on a common scale (Stevens 1957).
In all conditions, participants estimated their available financial slack in the current
month, 3 months later, 12 months later, and 36 months later using a single item that asked
participants to rate their available spare money on the same 11-point scale used Studies 1 - 3.
There were two different measurement conditions for income and expense. In the Ratings Scale
condition, participants rated their monthly income and expense change in 3, 12, and 36 months
on an 11-point scale. In the Modulus condition, participants were asked to imagine a pile of 100
poker chips that represent the size of their current income. Participants were then instructed to
write down how many poker chips would represent their income in 3, 12, and 36 months.
Participants used the same procedure to estimate their change in expenses over this time. Finally,
EXPENSE NEGLECT
20
participants were asked to set their current income at 100 poker chips and were asked to tell us
how many poker chips represented their current expenses. This was used to scale current income
to current expenses. For brevity, we present the results for the 12 months’ time frame below. The
results for 3 and 36 months are presented in Web Appendix A.
Participants in the Modulus condition showed significant expense neglect. We log
transformed the income and expenses change measures (ln +1), which greatly increases the fit of
our full model (R2 = .20) compared to using raw values (R2 = .11).
We regressed slack change on ln(income change +1) and ln(expense change +1). The
coefficient on income change (β = 1.02, SE = .19, t(116) = 5.32, p < .001) is 3.9 times the size of
the coefficient on expense change (β = -.26, SE = .21, t(116) = -1.23, p = .22). Further, the full
model (R2 = .20) that allows the income and expense parameters to vary explains a significantly
greater amount of variance than the constrained model where income and expenses have equal
and opposite signs (R2 = .12; F(1, 116) = 11.47, p = .001).
The reader may notice a change in the predictions of the normative model with equal
weights on income and expense when calculating coefficients based on log transformations.
With log transformations, the coefficient on the income and expense change now represents a
percentage change rather than an additive change. As a result, the coefficient compares increases
in percentage changes of income and expenses. However, if average current income is greater
than average current expenses—as is the case in the present study—then the log model suggests
that the coefficient on expenses normatively should be greater than the coefficient on income.
To illustrate, consider someone who currently makes $2,000 in income and spends $1,000 in
expenses per month. Now, imagine that this person’s income increases by 10% to $2,200, which
corresponds to a $200 increase in slack. In order to match this increase in income with a decrease
EXPENSE NEGLECT
21
in expenses, this person’s expenses would need to decrease by 20% so they are spending $200
less per month. In this case, a 10% increase in income is equal in weight to a 20% decrease in
expenses. As a result, the equal weights model presents a conservative test of expense neglect.
Expense neglect was again found in the Ratings Scale condition. We ran a regression
with income change and expense change to predict slack change. The coefficient on income
change (β = .92, SE = .12, t(122) = 7.82, p < .001) is 1.5 times the size of the coefficient on
expense change (β = -.60, SE = .12, t(122) = -5.10, p < . 001). Further, the full model (R2 = .37)
that allows the income and expense parameters to vary explains a significantly greater amount of
variance than the constrained model (R2 = .34; F(1, 122) = 5.31, p = .022).
In sum, study 4 shows that our results hold when using a common scale to measure
income and expense change. Note that even though the modulus uses a common scale, the
Ratings Scale condition still does a better job of explaining the observed data (R2Modulus = .20 vs
R2RatingsScale = .34), a result that holds across all time periods. This supports the notion that
participants are not using the response scales differently when evaluating income and expenses,
and ratings scales are sufficient for estimating perceptions about income and expenses.2
Nonetheless, in the next study, we ask participants to estimate their income and expenses using
dollars, another common scale that may be more natural for participants to evaluate.
Study 5: Measuring Income and Expenses in Dollars Does Not Eliminate Expense Neglect
2
One might conjecture that ratings scale produces a better model fit because slack is also measured using the same
scale. We conducted a replicate of the present study with university students (N = 208), but used a modulus to
measure financial slack instead of a ratings scale. We find that using a ratings scale to measure income and expenses
fits the data better across 3 months (R2Modulus = .19 vs R2RatingsScale = .27) and 12 months (R2Modulus = .07 vs R2RatingsScale
= .10), though not for 36 months (R2Modulus = .14 vs R2RatingsScale = .12). Thus the appropriateness of using ratings
scales to measure income and expenses appears robust to different measures of slack. Furthermore, there is no
evidence that this measurement scale drives the observed results.
EXPENSE NEGLECT
22
In Study 5 (N = 121), participants estimated their income and expenses in dollars.
Participants were asked to answer four separate focal questions asking participants to estimate
their income and expenses next month and in 24 months in dollars. These questions were
embedded in a larger set of estimates, so that participants would not reduce the task to a simple
algebra problem. Participants then evaluated their financial slack using the three-item measure
used in study 3 (α = .82).
As in the previous study, we logged transformed the income change and expense change
measures (ln + 1), which greatly improves the fit of the full model (R2 = .10) compared to using
raw values (R2 = .02). We ran a regression with income change and expense change to predict
slack change, and find that the coefficient on income change (β = .63, SE = .24, t(120) = 2.58, p
= .009) is significant, whereas the coefficient on expense change (β = .22, SE = .35, t(120) =
0.64, p = .52) is positive and not significant. Further, the full model (R2 = .10) that allows the
income and expense parameters to vary explains a significantly greater amount of variance than
the constrained model (R2 = .04; F(1, 118) = 7.61, p = .007), again replicating expense neglect.
As noted in Study 4, given that participants’ current income is greater than current expenses, the
equal weighting model represents a conservative test of expense neglect when using log
transformed predictors.
Together, studies 2 – 5 replicates expense neglect, and shows that this bias is robust to a
range of different operational definitions and measures of financial slack, income, and expenses.
We are therefore confident that the effect is neither a measurement nor a statistical artifact. In the
following studies, we investigate possible mechanism for the effect.
STUDY 6: UNCERTAINTY DOES NOT EXPLAIN EXPENSE NEGLECT
EXPENSE NEGLECT
23
The present study examines whether expense neglect is due to differing levels of
confidence about future income and expenses. If participants feel that their estimates about their
income are more certain than their estimates about their expenses, then they may rely more
heavily on income than expenses when estimating future financial slack.3 To test this, we directly
ask participants to report their level of confidence in their estimates.
Method
We recruited 301 participants (32.6% female; mean age = 31.8; median household
income = $40 to $50k) via Mechanical Turk to participate in a study about personal finances.
Participants evaluated their financial slack, income, and expenses in the next month and in 24
months from the present. After each estimate, participants were asked, “How confident are you
in your response?” on a scale ranging from 1 = “Not at all confident” to 11 = “Extremely
confident.” To measure financial slack, participants were asked the “spare money” and
“likelihood of repayment” questions. The change scores for these items were combined to create
a two-item slack change measure (r = .49). Financial constraint was not rated in this study.
Results & Discussion
Slack, Income and Expense Change: Relative to next month, participants expected their
slack (M = 1.87, SD = 2.25), income (M = 1.55, SD = 1.99), and expenses (M = 0.76, SD = 1.75)
to grow over the next two years (paired ts > 7.5, ps < .001).
3
If participants were highly uncertain in their expense estimates, then across our studies, the coefficients on expense
change should also likely be noisier (i.e., have a larger standard error) than the coefficients on the income change.
We do not find this, suggesting that uncertainty over expenses does not explain expense neglect. Nonetheless, we
directly test this account in the present study.
EXPENSE NEGLECT
24
Predictions of Slack Change: We again replicate expense neglect. We ran a regression
with income change and expense change to predict slack change, and find that the coefficient on
income change (β = .62, SE = .06, t(299) = 10.62, p < .001) is 3.1 times the size of the coefficient
on expense change (β = -.20, SE = .07, t(299) = -2.97, p = .003). Further, the full model (R2
= .28) that allows the income and expense parameters to vary explains more variance than the
constrained model (R2 = .20; F(1, 299) = 32.88, p < .001).
Confidence Ratings: A 2 x 2 repeated measures ANOVA showed a significant interaction
such that relative confidence in income vs. expense ratings varied by time period (next month vs
24 months), F(1, 300) = 11.26, p < .001. Planned contrasts show that next month, participants
were slightly more confident about their income (M = 9.34, SD = 1.93) than their expenses (M =
9.14, SD = 1.83), paired t(300) = 1.77, p = .077. However, predicting 24 months out, they were
actually significantly less confident about their income (M = 7.83, SD = 2.42) than their expenses
(M = 8.09, SD = 2.39), paired t(300) = -2.27, p = .024.
In sum, study 6 shows that expense neglect is not due to differing levels of confidence
between income and expenses. While the effect is small, participants were actually slightly more
confident in their own expense estimates than their own income estimates in two years. As a
result, it is unlikely that differences in confidence levels are causing people to neglect expenses
when forecasting their future financial slack.
STUDY 7: CATEGORY FLEXIBILITY DOES NOT EXPLAIN EXPENSE NEGLECT
Study 7 examines whether expense neglect can be explained by differing beliefs in the
flexibility of future income and expenses. Consumers may expect that their expenses will
increase in the future, but also consider some of these expenses to be discretionary and capable
EXPENSE NEGLECT
25
of being cut if need be. If so, they may discount flexible expenses when considering future
financial slack. For example, a consumer that expects to purchase a cable package once he
receives a raise may also consider this to be a flexible expense: if he receives an unexpected bill
in the future, he would be able to pay this bill by canceling his cable package and freeing up
spare money. As a result, he may indicate that his income, expenses, and spare money will all
increase in the future, leading to a pattern of results that will look like expense neglect. If
category flexibility can explain expense neglect, then we should see that participants who expect
their expenses to be more flexible in the future place less weight on expenses than those who
expect to have less flexible expenses in the future.
Method
We recruited 500 participants (43.2% female; mean age = 33.2; median household
income = $30 to $40k) via Mechanical Turk to participate in a study about personal finances.
Participants were asked about financial slack, income, and expenses in the next month and in 24
months from the present. Financial slack was measured using only the “likelihood of repayment”
item. After each estimate of their income and expenses, participants were asked the following:
“Some people's income is [expenses are] highly flexible, and they have the ability to quickly
increase or decrease their income [expenses] if need be. Other people's income is [expenses are]
not at all flexible, and they would find it impossible to quickly adjust their income [expenses].”
Participants then evaluated their income/expense flexibility on an 11-point scale ranging from 1
= “Not at all flexible” to 11 = “Extremely flexible.”
Results & Discussion
EXPENSE NEGLECT
26
Slack, Income and Expense Change: Relative to next month, participants expected their
slack (M = 1.63, SD = 2.81), income (M = 1.09, SD = 1.76), and expenses (M = 0.22, SD = 2.09)
to grow over the next two years (paired ts > 2.35, ps < .02).
Predictions of Slack Change: We again replicate expense neglect. We ran a regression
with income change and expense change to predict slack change, and find that the coefficient on
income change (β = .43, SE = .07, t(497) = 6.09, p < .001) is positive and significant, while the
coefficient on expense change (β = .04, SE = .06, t(497) = 0.63, p = .53) is not significantly
different from zero. Further, the full model (R2 = .08) that allows the income and expense
parameters to vary explains significantly more variance than the constrained model where
income and expenses have equal and opposite signs (R2 = .02; F(1, 497) = 33.32, p < .001).
Category Flexibility: Next, we tested to see if the extent to which income and expenses
are considered to be flexible affects the weight placed on expenses. To do so, we first took
change scores between extent to which participants expected their income flexibility and
expenses flexibility to change over time, and ran the following regression:
(5) Slack Change =
β0 + β1(Income Changei) + β2(Expense Changei) +
β3(Income Flexibility Changei) + β4(Expense Flexibility Changei) +
β5(Income Changei*Income Flexibility Changei) +
β6(Expense Changei*Expense Flexibility Changei) + i
Both income flexibility change (M = 0.74, SD = 2.00) and expense flexibility change (M = 0.52,
SD = 2.09) were mean centered, whereas income change and expense change were coded such
that 0 meant no change.
EXPENSE NEGLECT
27
Table 2 displays the regression results, and shows that category flexibility does not affect
the weighting of either income or expenses. Both the income change by income flexibility
change interaction and the expense change by expense flexibility change interaction are not
significant. Thus, those expecting greater income or expense flexibility in the future did not
weigh these inputs any differently than those reporting less income or expense flexibility.
---Insert Table 2 about here---
Interestingly, we do find a significant positive relationship between increased expense
flexibility and increased slack, suggesting that the more participants believe that their expenses
will be flexible in the future, the more they expect to have greater financial slack in the future.
However, participants are still insensitive to their overall level of expenses: controlling for
flexibility, the coefficient on expense change remains not significantly different from zero.
In sum, study 7 shows that beliefs regarding the flexibility of expenses cannot explain
expense neglect. Although participants who expected to have greater expense flexibility in the
future reported having greater slack in the future, there was no relationship between expense
flexibility change and expense change.
STUDY 8: OPTIMISM DOES NOT EXPLAIN EXPENSE NEGLECT
Study 8 examines whether expense neglect can be explained by generalized optimism. In
particular, it may be that people who are optimistic and view the future favorably are also more
likely to ignore expenses when forecasting their future slack.
EXPENSE NEGLECT
28
Method
We recruited 1,004 participants4 (39.0% female; mean age = 30.1; median household
income $40k - $50k) via Mechanical Turk. Participants first rated their available financial slack
next month and in two years from the present using the same three-item measure developed in
Study 4 ( = .81). Participants then evaluated their expected monthly income and expenses next
month and in two years from the present. The income, expense, and slack measures were all
rated using 11-point scales. Finally, participants were given the Life Orientation Test – Revised
(LOT-R) as a measure of generalized optimism (Scheier, Carver, and Bridges 1994). The scores
for the LOT-R range from 0 (extreme pessimist) to 24 (extreme optimist).
Results & Discussion
Slack, Income and Expense Change: Relative to next month, participants expected their
slack (M = 2.55, SD = 2.54), income (M = 1.94, SD = 2.19), and expenses (M = 0.20, SD = 2.15)
to grow over the next two years (paired ts > 2.87, ps < .005).
Predictions of Slack Growth: We again find evidence for expense neglect. We ran a
regression with income change and expense change to predict slack change, and find that the
coefficient on income change (β = .62, SE = .03, t(1001) = 19.22, p < .001) is 3.0 times the size
of the coefficient on expense change (β = -.21, SE = .03, t(1001) = -6.56, p < .001). Further, the
full model (R2 = .27) that allows the income and expense parameters to vary explains
significantly more variance than the constrained model where income and expenses have equal
and opposite signs (R2 = .20; F(1, 1001) = 99.82, p < .001).
4
We recruited this number of participants to match the sample recruited in Study 9, which was run prior to the
present study. These sample sizes were matched to ensure that we would not under-power the present study.
EXPENSE NEGLECT
29
Optimism: Participants on average were slightly above the mid-point of the LOT-R scale
(M = 13.26, SD = 5.19). We first checked to see if optimists viewed their future finances more
favorably than pessimists. More optimistic respondents expected greater future income (r = .22,
p < .001), smaller future expenses (r = -.11, p < .001), and greater future slack (r = .37, p < .001).
More optimistic participants also expected greater future income growth (r = .08, p = .016), and
greater future slack growth (r = .10, p = .002), but did not expect any differences in expense
growth (r = .03, p = .29). These results attest to the validity of the LOT-R scale.
Next we tested to see if the tendency for income to be weighted more than expenses is
accounted for by one’s level of optimism. To do so, we ran the following regression, with
Income Change and Expense Change centered at 0, and LOT-R mean centered:
(6) Slack Change =
β0 + β1(Income Changei) + β2(Expense Changei) + β3(LOT-Ri) +
β4(Income Changei*LOT-Ri) + β5(Expense Changei*LOT-Ri) + i
Table 3 shows the full regression results. Notably, both the income change by LOT-R interaction
and the expense change by LOT-R interaction are not significant, while we replicate expense
neglect controlling for optimism. In other words, regardless of their level of optimism,
consumers equally relied on their own income change forecasts to predict slack change and they
equally under-weighed the role of expense change in affecting slack change.
---Insert Table 3 about here---
EXPENSE NEGLECT
30
In sum, study 7 shows that generalized optimism does not explain expense neglect. While
those higher on optimism were more likely to project to have a more favorable financial future,
they were no more likely to ignore expenses when estimating their future financial slack.
STUDY 9: SPENDTHRIFTS ARE MORE PRONE TO EXPENSE NEGLECT THAN
TIGHTWADS
In study 9, we examine whether consumers who are chronically attuned to expenses
would be less likely to display expense neglect. Previous research shows that tightwads
experience greater pain of paying and are less likely to neglect opportunity costs than
spendthrifts (Rick, Cryder, and Loewenstein 2008; Frederick et al. 2009). We expect that
tightwads’ sensitivity to costs would cause them to be more likely to attend to and account for
expenses when estimating future financial slack. An alternative approach to examining the role
of attention would be to experimentally manipulate the level of attention to income and expenses.
We will discuss two attempts to experimentally manipulation attention in the discussion section
of this study.
Method
A power analysis based on a pre-test revealed an estimated sample of 942 participants for
power of .80 (see Web Appendix A for details). As a result, we recruited 1,000 participants
(31.5% female; mean age = 27.9; median household income $40k - $50k) to participate in this
study via Mechanical Turk. Participants rated their available financial slack next month and in
two years from the present using the same three-item measure developed in Study 4 ( = .78).
Participants also evaluated their expected monthly income and expenses next month and in two
EXPENSE NEGLECT
31
years from now. Income, expense, and slack measures were all rated using 11-point scales.
Finally, participants were given the four-item Tightwad-Spendthrift (T-S) scale (Rick, Cryder,
and Loewenstein 2009). Scores for the T-S scale range from 4 (extreme tightwad) to 26 (extreme
spendthrift).
Results
Slack, Income and Expense Change: Participants expected their slack (M = 2.23, SD =
2.56), income (M = 1.98, SD = 2.13), and expenses (M = 0.48, SD = 2.27) to grow in 2 years
relative to next month (paired ts > 6.74, ps < .001).
Prediction of Slack Growth: We again find evidence for expense neglect. We ran a
regression with income change and expense change to predict slack change, and find that the
coefficient on income change (β = .56, SE = .03, t(997) = 16.51, p < .001) is 2.8 times the size of
the coefficient on expense change (β = -.20, SE = .03, t(997) = -6.43, p < .001). Further, the full
model (R2 = .23) that allows the income change and expense change parameters to vary explains
significantly more variance than the constrained model where income and expenses have equal
and opposite signs (R2 = .18; F(1, 997) = 64.64, p < .001).
Tightwad-Spendthrift Scale: We then tested our main hypothesis, which is that tightwads
are more likely to account for expenses when forecasting slack change than spendthrifts.
Specifically, we ran the following regression:
(7) Slack Change =
β0 + β1(Income Changei) + β2(Expense Changei) + β3(T-Si) +
β4(Income Changei*T-Si) + β5(Expense Changei*T-Si) + i
EXPENSE NEGLECT
32
We found a significant expense change by T-S interaction (β = .013, SE = .006, t(994) = 2.13, p
=.034), in the predicted direction: the lower participants scored on the T-S scale (the more they
were considered tightwads), the more likely they were to account for expenses when estimating
their future financial slack. However, the income change by T-S interaction was not significant
(β = .001, SE = .007, t(994) = .22, p =.83), suggesting that tightwads and spendthrifts weigh
income equally. There was also a simple effect of T-S on slack change (β = .081, SE = .021,
t(994) = 3.91, p < .001), such that spendthrifts report estimating greater slack change in the
future than tightwads when income change and expense change = 0.
Next, we conducted a floodlight analysis (Aiken and West 1990; Spiller, Fitzsimons,
Lynch, and McClelland 2013) to see how the coefficients on income change and expenses
change at specific levels of the T-S scale. Figure 2 displays these results.
---Insert Figure 2 about here---
Several points emerge. First, the coefficient on income is significant for all values of the
T-S scale, as shown by the fact that 95% confidence bands exclude zero throughout the graph.
Second, the coefficient on expenses is significant for all T-S values beyond the Johnson-Neyman
point of 21.6—only extreme spendthrifts entirely neglect expenses. More importantly, when we
centered the scale on extreme tightwads (T-S = 4), the coefficient on income change (β = .54, SE
= .07, t(994) = 7.70, p < .001) is only 1.6 times the size of expense change (β = -.33, SE = .07,
t(994) = -4.58, p < .001). However, when we centered the scale on extreme spendthrifts (T-S =
26), the coefficient on income change (β = .57, SE = .09, t(994) = 6.40, p < .001) is now 14.3
times the size of expense change (β = -.04, SE = .08, t(994) = -.56, p = .58).
EXPENSE NEGLECT
33
Discussion
Two results are critical to interpret the analysis above and the implications for expense
neglect. First, the weight of income change does not differ significantly by score on the T-S scale,
as evidenced by the lack of an income change by T-S interaction. However, the weight of
expense change does differ by the T-S score, as evidenced by a significant interaction expense
change by T-S interaction. This is consistent with an interpretation that the difference between
tightwads and spendthrifts is about sensitivity to expenses, and that tightwads are more likely to
attend to expenses when forecasting their future financial slack. Second, the coefficient on
income change is greater in magnitude than the coefficient on expense change across the entire
range of T-S scale (although to varying degree). That is, even extreme tightwads weigh income
greater than expenses, attesting to the robustness of the effect.
One question that arises when considering testing sensitivity towards expenses is whether
a direct manipulation that increases attention towards expenses would “de-bias” participants. For
example, in two unreported studies (see Web Appendix B for full details) we attempted to
systematically manipulate participants’ attention to the various inputs into slack. While we found
some evidence that our attention manipulations increased the weight placed on expenses, we did
not find this result to be robust. Given the reliability of expense neglect, we expect that this
inconsistency is due to the fact that these assessments are deeply ingrained and not easily
overridden using the type of manipulations we employed in an experimental session.
Next, we present a meta-analysis of our entire file drawer of 25 studies to further evaluate
the robustness of expense neglect. To gain further insight about possible temporal variations in
the attention towards income and expenses we additionally examine how consumers weigh
EXPENSE NEGLECT
34
income and expenses separately in the immediate versus distant future. Consistent with an
attention account, we expect to find evidence that consumers decrease their attention towards
expenses with time.
META-ANALYSIS
We conducted a series of meta-analyses to test whether expense neglect persists across
our entire file drawer of 25 studies and 7,214 participants (see Web Appendix C for an overview
of studies). In the first part, we examine how participants weight income change and expense
change when predicting slack change, consistent with our analyses conducted thus far.
In the second part, we examine the subset of our studies that ask participants to make
estimates about their income, expenses, and slack at a given point in time (e.g., how much
expenses they expect to have next month [in two years from now]), and exclude studies that
directly ask participants to make estimates regarding changes over time (e.g., how much they
expect their expenses to change in two years from now). This allows us to directly compare how
much weight participants place in income and expenses in the immediate future separately from
the weight that they place on income and expenses in the distant future.
Estimating Slack Change over Time
We first examined the extent to which participants weigh income change relative to
expense change when estimating slack change. Each study contributed three partial correlations
into the meta-analysis, which were all converted into Fisher’s Z:

2
2
The semi-partial r of the model comparison: Semi-partial r = √𝑅𝐹𝑢𝑙𝑙
− 𝑅𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑒𝑑
Full Model:
Slack Changei = β0 + β1Income Changei + β2Expense Changei + i
EXPENSE NEGLECT
Constrained Model:
35
Slack Changei = β0 + β1(Income Changei - Expense Changei) + i

The partial r between Income Change and Slack Change from the full model.

The partial r between Expense Change and Slack Change from the full model.
We computed Fisher’s Z for each semi-partial correlation from each study and calculated
a weighted mean Z, using inverse variance weights to assign more weight to studies with larger
samples (Lipsey and Wilson 2001). We converted Fisher’s Z values back to r for presentation in
what follows, for ease of interpretation.
Table 3 presents an overview of the meta-analysis results, and Web Appendix C contains
the full data for the meta-analysis. Overall, we find that across studies, participants weigh
income change more heavily than expense change when predicting future slack. The mean effect
size of the Model Comparison semi-partial r is .210 (95% CI = 0.187, 0.232). The mean effect
size of the Income Change partial r is 0.398 (95% CI = 0.378, 0.417) and the Expenses Change
partial r is -0.146 (95% CI = -0.169, -0.123), which corresponds to assigning 2.7 times more
weight to standardized income change than standardized expense change.
We supplemented this analysis by examining whether different operationalizations across
our studies contributed to variation in our results. Results show significant variation in effect
sizes across studies (I2 Model Comparison = .64). This is not surprising since we intentionally
used a variety of different operationalizations across our studies. We then coded each study for
the variables that we expected to cause the variation in effect sizes, including: 1) the log of the
distance from the present to the future time period; 2) whether income and expenses were
measured as static (estimates were taken of both the immediate and distant future, then we
calculated the difference between these two periods) or growth measures (a direct estimate of
EXPENSE NEGLECT
36
how much income and expense were expected to grow); 3) whether income and expenses were
measured using a modulus or (a little to a lot) rating scale; and 4) whether the present time period
refers to the current period (this month) or the next period (next month).
We then conducted meta-analysis regressions to test whether these predicted factors
uniquely contributed to the effect size variation. We ran a fixed effect model, whereby we
regressed the effect size of the model comparison on the four factors listed above. Except for the
distance-from-present factor (β = .10, p = .06), no other factors predicted the effect size of the
model comparison (ps > .46). Due to high correlations among the four predictors listed above
(.15 < rs < .85), we also ran four separate random effect models to predict the effect size of the
model comparison. Each regression used only one of the four predictors in each model. The only
significant model was again the one that included just the distance-from-present factor (β = .10, p
= .05). None of the other predictors approached statistical significance (ps > .14). Thus, the
further in time that participants evaluate their finances, the more they exhibit expense neglect.
We further analyzed the relationship between participants’ expectations of their future
income and expense growth. Across studies, we found a moderate correlation between expected
income and expense growth (r = .219; 95% CI [.197, .241]). In other words, the more
participants expected their income to grow, the more they expected their expenses to grow.
Further, participants expected their income to grow to a greater extent than their expenses over
the same time period (d = .698; 95% CI [.646, .750]).
Estimating Total Slack at Different Points in Time
The preceding analysis shows that when examining slack growth, participants place on
average 2.7 times the weight on standardized income than standardized expenses. This reflects
EXPENSE NEGLECT
37
the analyses reported throughout the paper. We supplement this analysis with an additional
examination comparing participants’ estimates of their slack at a given point of time to provide
converging evidence in support of our findings and interpretation. Thus, instead of estimating
change scores, we estimate slack for individual i at specific points in time, as represented in
equations 8a and 8b:
(8a) Immediate Future Slack = β0 + β1(Imm. Future Incomei) + β2(Imm. Future Expensesi) + i
(8b) Distant Future Slack = β0 + β1(Dist. Future Incomei) + β2(Dist. Future Expensesi) + i
If participants are neglecting expenses at all points in time equally, then the weight on Expenses
(β2) should be similar in the distant future than the immediate future. However, if expenses in the
present are more salient then expenses in the future, then β2 should be significantly smaller in the
distant future than the immediate future.
We examined all of our studies that asked for separate ratings of immediate future and
distant future values of the three constructs income, expenses, and slack—omitting studies where
we only asked participants to rate growth directly. These measures were assessed in 14 studies
(N = 4,607). We estimated equation 8a & 8b, and extracted the partial r for income and expenses
both in the immediate and distant future. We then conducted the same meta-analysis as described
above, weighting the Fisher’s Z of each partial r by inverse variance, and finally converted the
weighted mean Fisher’s Z values back to r.
Table 4 presents an overview of results of the meta-analysis, and Web Appendix C
contains details of the individual studies for the meta-analysis. Results show that for the
immediate future, the mean effect size of the partial r for income is 0.503 (95% CI = 0.481,
EXPENSE NEGLECT
38
0.525), and for expenses is -0.311 (95% CI = -0.337, -0.284). For the distant future, the mean
effect size of the partial r for income is 0.564 (95% CI = 0.544, 0.584), and of expenses is -0.250
(95% CI = -0.277, -0.222). Thus, relative to the immediate future, the weight on income
increases significantly by 12.1% in the distant future and for expenses it decreases significantly
by 24.4%. Looking at it another way, the ratio of the weight placed on income relative to
expenses increases by 39.5%, from 1.62 in the immediate future to 2.26 in the distant future.
This provides further evidence that the further participants look into the future, the greater
weight that they place on income relative to expenses when estimating their financial slack.
Finally, we evaluated the model comparison measures for the immediate future and the
distant future separately. For the immediate future, the mean effect size of the model comparison
semi-partial r is .189 (95% CI = 0.160, 0.217), and for the distant future it is .229 (95% CI =
0.201, 0.257), which corresponds to a 21.1% increase in effect size.
---Insert Table 4 about here---
GENERAL DISCUSSION
This paper examines how consumers forecast their personal finances into the future, and
demonstrates robust evidence for expense neglect, which we define as the tendency to underweigh expected changes in expenses relative to expected changes in income when predicting
future financial slack. We show that expense neglect is not due to psychometric measurement
issues and cannot be explained by a lack of confidence in predictions about future expenses,
beliefs that future expenses are more flexible than current expenses, or a general optimism bias.
Rather, we argue that expense neglect is in part due to insufficient attention to expenses, and find
EXPENSE NEGLECT
39
that individual differences in sensitivity towards expenses matter. Those who are particularly
sensitive to expenses (tightwads) are more likely to account for expenses when estimating their
future financial slack than those who are insensitive to expenses (spendthrifts).
Further, a meta-analysis of our entire file drawer shows robust evidence in expense
neglect, and that expense neglect is stronger when evaluating time periods more distant in the
future. We supplement this analysis by looking at participants’ static estimates of their finances
at a given point in time. In these predictions, people place greater weight on income and less
weight on expenses in the distant future as compared to the immediate future. This provides
additional evidence that people are more attentive to expenses in the present and less attentive to
expenses in the future when estimating their future financial slack, which is consistent with an
attention mechanism.
Expense Neglect as a Focusing Illusion
The attention explanation we propose is similar to that of the focusing illusion in
affective forecasting (Kahneman, et al. 2006; Schkade and Kahneman 1998; Ubel et al. 2001),
and provides an appropriate analogy for our findings. Just as those from the Midwest focus too
much on the role of sunshine when thinking about life in California (and not enough about the
traffic), those who think about their finances may focus too much on the role of income and not
enough on the role of expenses. Consumers may fail to realize that earning extra money, and
then spending this money, will squeeze their available resources and leave them with less slack
than they originally anticipated.
Determinants of Financial Slack
EXPENSE NEGLECT
40
More broadly, the research presented here speaks to how consumers form subjective
expectations of financial slack. While we highlight expense neglect as our main finding, we
additionally find that consumers expect their financial slack to increase in the future, and they
expect it to increase more so in the distant future than the near future. We also find that optimists
and spendthrifts expect to have more slack in the future than pessimists and tightwads.
We additionally find that expected changes in expense flexibility matter, but expected
changes in income flexibility do not. Specifically, the more consumers expect their expenses to
become flexible in the future, the more they expect to have greater slack in the future. However,
we do not find a strong relationship between future income flexibility and slack change. Future
research can investigate why flexibility appears to matter more for expenses than income.
Conclusion
In conclusion, the research presented here helps understand one reason why people feel
financially constrained in the present, yet believe their future finances are more favorable. When
forecasting the future, they under-weigh the impact of the growth in expenses relative to the
growth in income on their future finances.
EXPENSE NEGLECT
41
REFERENCES
Aiken, Leona S. and Stephen G. West (1991), Multiple Regression: Testing and Interpreting
Interactions. Newbury Park, CA: Sage Publications.
Anderson, Norman H. (1982), Methods of Information Integration Theory. Vol. 2. New York:
Academic Press.
Booth Sweeney, Linda and John D. Sterman (2000), “Bathtub Dynamics: Initial Results of a
Systems Thinking Inventory,” System Dynamics Review, 16 (4), 249-86.
Bricker, Jesse et al. (2014), “Changes in U.S. Family Finances from 2010 to 2013: Evidence
from the Survey of Consumer Finances,” Federal Reserve Bulletin, 100 (4), September,
available at http://www.federalreserve.gov/pubs/bulletin/2014/pdf/scf14.pdf
Buehler, Roger, Dale Griffin, and Michael Ross (1994), “Exploring the “Planning Fallacy”: Why
People Underestimate Their Task Completion Times,” Journal of Personality and Social
Psychology, 67 (3), 366-81.
Cohen, Jacob, Patricia Cohen, Stephen G. West, and Leona S. Aiken (2003), Applied Multiple
Regression/Correlation Analysis for the Behavior Sciences, 3d ed. Mahwah, NJ:
Lawrence Erlbaum Associates.
Cook, Thomas K. and Donald T. Campbell (1979), Quasi-Experimentation: Design and Analysis
Issues for Field Settings. Chicago: Rand McNally.
Cronin, Matthew A., and Cleotilde Gonzalez (2007). “Understanding the Building Blocks of
Dynamic Systems,” System Dynamics Review 23, (1),) 1-17.
_____, Cleotilde Gonzalez, and John D. Sterman (2009). “Why Don’t Well-Educated Adults
Understand Accumulation? A Challenge to Researchers, Educators, and
Citizens,” Organizational Behavior and Human Decision Processes 108, (1): 116-30.
EXPENSE NEGLECT
42
Frederick, Shane, Nathan Novemsky, Jing Wang, Ravi Dhar, and Stephen Nowlis (2009),
“Opportunity Cost Neglect,” Journal of Consumer Research 36 (4), 553-61.
Kahneman, Daniel, Jack L. Knetsch, and Richard H. Thaler (1991), “Anomalies: The
Endowment Effect, Loss Aversion, and Status Quo Bias,” Journal of Economic
Perspectives, 5 (1), 193-206.
_____, Alan B. Krueger, David Schkade, Norbert Schwarz, and Arthur A. Stone (2006), “Would
You Be Happier If You Were Richer? A Focusing Illusion,” Science, 312 (5782), 190810.
_____ and Amos Tversky (1979), “Intuitive Prediction: Biases and Corrective Procedures,”
TIMS Studies in Management Science, 12, 313-27
Kivetz, Ran and Anat Keinan (2006), "Repenting Hyperopia: An Analysis of Self-Control
Regrets," Journal of Consumer Research, 33 (September), 273-82.
Lipsey, Mark W. and David B. Wilson (2001), Practical Meta-Analysis. Sage,Thousand Oaks,
CA.
Lynch Jr., John G., Dipankar Chakravarti, and Anusree Mitra (1991), “Contrast Effects in
Consumer Judgments: Changes in Mental Representations or in the Anchoring of Rating
Scales?,” Journal of Consumer Research, 18 (December), 284-97.
_____ and Gal Zauberman (2006), “When Do You Want It? Time, Decisions, and Public
Policy,” Journal of Public Policy & Marketing, 25 (1), 67-78.
Peetz, Joanne and Roger Buehler (2009), “Is There a Budget Fallacy? The Role of Savings Goals
in the Prediction of Personal Spending,” Personality and Social Psychology Bulletin, 35
(12), 1579-91.
_____ and ______ (2012), “When Distance Pays Off: The Role of Construal Level in Spending
EXPENSE NEGLECT
43
Predictions,” Journal of Experimental Social Psychology, 48, 395-98.
Puri, Manju and David T. Robinson (2007), “Optimism and Economic Choice,” Journal of
Financial Economics, 86, 71–99.
Rick, Scott I., Cynthia E. Cryder, and George Loewenstein (2008), “Tightwads and
Spendthrifts,” Journal of Consumer Research, 34 (6), 767-82.
Scheier, Michael F., Charles S. Carver, and Michael W. Bridges (1994), “Distinguishing
Optimism from Neuroticism (and Trait Anxiety, Self-Mastery, and Self-Esteem): a
Reevaluation of the Life Orientation Test,” Journal of Personality and Social
Psychology, 67 (6), 1063-78.
Schkade, David A. and Daniel Kahneman (1998), “Does Living in California Make People
Happy? A Focusing Illusion in Judgments of Life Satisfaction,” Psychological Science, 9
(5), 340-46.
Shah, Anuj K., Sendhil Mullainathan, and Eldar Shafir (2012), “Some Consequences of Having
Too Little,” Science, 382, 682-85.
Sharma, Eesha and Adam L. Alter (2012), “Financial Deprivation Prompts Consumers to Seek
Scarce Goods,” Journal of Consumer Research, 39 (3), 545-60.
Slovic, Paul, Dale Griffin, and Amos Tversky (1990), “Compatibility Effects in Judgment and
Choice,” Insights in Decision Making: A Tribute to Hillel J. Einhorn: 5-27.
Spiller, Stephen A., Gavan J. Fitzsimons, John G. Lynch, Jr., Gary H. McClelland (2013),
“Spotlights, Floodlights, and the Magic Number Zero: Simple Effects Tests in Moderated
Regression,” Journal of Marketing Research, 50 (April), 277-88.
Sterman, John D. and Linda Booth Sweeney (2002), “Cloudy Skies: Assessing Public
Understanding of Global Warming,” System Dynamics Review 18 (2), 207-40.
EXPENSE NEGLECT
44
Stevens, Stanley S. (1957), “On the Psychophysical Law,” Psychological Review, 64 (3), 153-81.
Sussman, Abigail B. and Adam L. Alter (2012), “The Exception is the Rule: Underestimating and
Overspending on Exceptional Expenses,” Journal of Consumer Research, 39, 800-14.
Thaler, Richard H. and Shlomo Benartzi (2004), “Save More Tomorrow™: Using Behavioral
Economics to Increase Employee Saving,” Journal of Political Economy, 112 (S1), S16487.
Tversky, Amos, Shmuel Sattath, and Paul Slovic (1988), “Contingent Weighting in Judgment
and Choice," Psychological Review, 95 (3), 371-84.
Ubel, Peter A., George Loewenstein, John Hershey, Jonathan Baron, Tara Mohr, David A. Asch,
and Christopher Jepson (2001), "Do Nonpatients Underestimate the Quality of Life
Associated with Chronic Health Conditions Because of a Focusing Illusion?" Medical
Decision Making, 21 (3), 190-99.
Ülkümen, Gülden, Manoj Thomas, and Vicki G. Morwitz (2008), “Will I Spend More in 12
Months or a Year? The Effect of Ease of Estimation and Confidence on Budget
Estimates,” Journal of Consumer Research, 35, 245-56.
Zauberman, Gal and John G. Lynch Jr. (2005), “Resource Slack and Propensity to Discount
Delayed Investments of Time Versus Money," Journal of Experimental Psychology:
General, 134 (1), 23-37.
EXPENSE NEGLECT
45
Table 1: Breakdown of Study 1 Results by Group
3 Months
Regression Coefficients
Income Expense
Growth Growth
Group
N
Intercept
Employed
150
0.29
(0.16)
0.30***
(0.09)
0.60***
(0.17)
Unemployed 150
12 Months
Model
Comparison
Regression Coefficients
Intercept
Income Expense
Growth Growth
F
p
0.12
(0.11)
9.74
.002
0.79*** 0.44***
(0.23)
(0.10)
0.13
(0.07)
0.05
(0.08)
3.15
.078
1.28*** 0.44***
(0.26)
(0.10)
Model
Comparison
F
p
-0.03
(0.12)
9.21
.003
-0.09
(0.10)
7.00
.009
Students
216
0.47**
(0.18)
0.57***
(0.12)
-0.08
(0.11)
12.97 <.001
0.70*
(0.31)
0.50***
(0.11)
0.005
(0.10)
15.48 <.001
Executives
64
0.35
(0.22)
0.10
(0.14)
0.09
(0.17)
1.23
0.41
(0.41)
0.45*
(0.19)
-0.09
(0.21)
2.37
.27
.13
Notes: The model comparison test examines the difference of fit between the full model where the weights of income
and expenses are free to vary to the constrained model where income and expenses are set to have equal and opposite
signs. Parentheses indicate standard errors. * p < .05, ** p < .01, *** p < .001
EXPENSE NEGLECT
46
EXPENSE NEGLECT
47
EXPENSE NEGLECT
48
Table 4: Summary of Meta-Analyses
Slack Assessment
Web
App.
Table #
N
Weighted Mean
Weighted
SE
Effect Size
Mean Effect
[Fisher's Z]
[Fisher's Z]
Size [r ]
95% CI [r ]
Slack Change
Model Comparison
4
7,214
.213
.012
.210
[.187, .232]
Slack Change
Partial Effect of Income
5
7,214
.421
.012
.398
[.378, .417]
Slack Change
Partial Effect of Expenses
5
7,214
-.147
.012
-.146
[-.169, -.123]
Immediate Future
Model Comparison
6
4,607
.191
.015
.189
[.160, .217]
Immediate Future
Partial Effect of Income
7
4,607
.554
.015
.503
[.481, .525]
Immediate Future
Partial Effect of Expenses
7
4,607
-.321
.015
-.311
[-.337, -.284]
Distant Future
Model Comparison
8
4,607
.233
.015
.229
[.201, .257]
Distant Future
Partial Effect of Income
9
4,607
.639
.015
.564
[.544, .584]
Distant Future
Partial Effect of Expenses
9
4,607
-.255
.015
-.250
[-.277, -.222]
Note: Slack Change referes to the anaylsis conducted on income change and expense change predicting slack change. The Immediate
Future and the Distant Future assessments refer to just the studies that evaluate slack at a given point in time, as represented in equations
8a and 8b, respectively. See Web Appendix C for details regarding each meta-analysis.
EXPENSE NEGLECT
49
Expected Slack Change
Expected Change
3.0
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
Employed
Unemployed
3 Months
Students
Executives
12 Months
Figure 1a: Expected slack change for the four samples taken in Study 1 as calculated from a
change score. Participants expected that their financial slack will grow over time, and will grow
more in the long term than the short term. Error bars represent 95% confidence intervals.
Expected Change in Income & Expenses
Expected Change
3.0
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
Employed
Unemployed
Students
Executives
3 Mo. Income Change
3 Mo. Expense Change
12 Mo. Income Change
12 Mo. Expense Change
Figure 1b: Expected income and expense changes for the four samples taken in Study 1.
Participants expected that their incomes and expenses will grow over time, and both are expected
to grow more in the long term than the near term. Error bars represent 95% confidence intervals.
EXPENSE NEGLECT
50
Income Change
Expense Change
1.0
Slope of the Effect on Slack Change
0.8
0.6
0.4
0.2
0.0
4
6
8
10
12
14
16
18
20
22
24
26
-0.2
-0.4
Johnson-Neyman
point = 21.53
-0.6
-0.8
-1.0
Extreme
Tightwads
Extreme
Spendthrifts
Figure 2: Floodlight analysis of the slope of effect of Expense Change (black lines) and Income
Change (grey lines) on Slack as a function of Tightwad-Spendthrift scores (4 = extreme tightwad,
26 = extreme spendthrift). The dotted vertical like corresponds to the Johnson-Neyman point of
21.58 whereby the slope of the expense curve becomes significantly different from zero. Error
bars represent 95% confidence intervals.
EXPENSE NEGLECT – WEB APPENDIX
1
WEB APPENDIX
Expense Neglect in Forecasting Personal Finances
WEB APPENDIX A
ADDITIONAL METHODS AND RESULTS
STUDY 2: ESTIMATING FUTURE FINANCIAL CONSTRAINT DOES NOT INCREASE
THE WEIGHT PLACED ON EXPENSES
Method
We recruited 302 participants (35.8% female; mean age = 31.3; median household
income = $40 to $50k) via Mechanical Turk to participate in a study about their personal
finances. Participants were randomly assigned into one of two response scale conditions. In the
Spare Money condition, participants evaluated the amount of available spare money they
expected in their monthly budget next month and in a month two years from the present (on an
11-point scale anchored by 1 = “Very little available money” to 11 = “Lots of available money”).
In the Financial Constraint condition, participants evaluated the amount of financial constraint
they expect to have in their budgets over the same time periods (on an 11-point scale anchored
by 1 = “Very little constraint” and 11 = “Lots of constraint.”).
Results
Predictions of Slack Growth: Pooling across response scale conditions, participants
expected their slack (M = 1.17, SD = 2.37), income (M = 1.67, SD = 2.11), and expenses (M =
EXPENSE NEGLECT – WEB APPENDIX
2
0.85, SD = 2.24) all to grow in the future. All three increases were significantly greater than 0
(one-sample ts > 6.6, ps < .001).
STUDY 3: MULTI-ITEM MEASURES OF FINANCIAL SLACK DOES NOT ELIMINATE
EXPENSE NEGLECT
Method
We recruited 74 participants (21.6% female; mean age = 25.2; median household income
= $30 to $40k) via Mechanical Turk to participant in a study about their personal finances.
Participants were asked to estimate their income, expenses, and financial slack next month and in
two years’ time.
Results
Participants expected their slack (M = 1.62, SD = 2.32), income (M = 1.81, SD = 1.81),
and expenses (M = 1.15, SD = 1.76) to grow in the future. All three increases were significantly
greater than 0 (one-sample ts > 6.0, ps < .001).
STUDY 4: RULING OUT RESPONSE LANGUAGE USING MAGNITUDE ESTIMATION
Method
We recruited 244 individuals (41.4% female; mean age = 31.8; median household income
= $30 to $40k) via Mechanical Turk to participate in a study about their personal finances.
Participants were allocated into either a Response Scale condition or a Modulus condition.
In all conditions, participants estimated their available financial slack in the current
month, 3 months later, 12 months later and 36 months later using a single item that asked
EXPENSE NEGLECT – WEB APPENDIX
3
participants to rate their available spare money on an 11-point scale. This study was run prior to
the development of the three-item measure used in the previous study.
In the Response Scale condition, participants rated their monthly income and expense
change in 3, 12, and 36 months on an 11-point scale anchored by -5 = “I expect my
income[expenses] to greatly decrease”; 0 = “I expect my income [expenses] will remain the
same”; and +5 “I expect my income[expenses] to greatly increase”.
In the Modulus condition, participants were asked to imagine a pile of 100 poker chips
that represent the size of their current income. Participants were then instructed to write down
how many poker chips would represent their income in 3, 12, and 36 months. Participants used
the same procedure to estimate their change in expenses over this time. Finally, participants were
asked to set their current income at 100 poker chips and were asked to tell us how many poker
chips represented their current expenses. This was used to scale current income to current
expenses.
Results
Modulus condition: Participants on average expected their slack to grow in 3 months (M
= 0.97, SD = 1.90; one-sample t(118) = 5.58, p < .001), 12 months (M = 1.73, SD = 2.32; onesample t(118) = 7.63, p < .001), and 36 months (M = 2.74, SD = 3.14; each t(118) > 5.57, p
< .001).
Participants’ current expenses were on average less than their current incomes (Median =
80%; Wilcoxon signed rank test Z = -5.52, p < .001). Participants expected their income to grow
in 3 months (Median Growth = 2%; Wilcoxon signed rank test Z = 5.86, p < .001), 12 months
(Median Growth = 25%; Wilcoxon signed rank test Z = 7.27, p < .001), and 36 months (Median
EXPENSE NEGLECT – WEB APPENDIX
4
Growth = 100%; Wilcoxon signed rank test Z = 8.71, p < .001). Participants did not expect their
expenses to grow significantly in 3 months (Median Growth = 0%; Wilcoxon signed rank test Z
= 1.16, p = .25), but did expect their expenses to grow in 12 months (Median Growth = 5%;
Wilcoxon signed rank test Z = 4.93, p < .001) and 36 months (Median Growth = 40%; Wilcoxon
signed rank test Z = 6.84, p < .001).
We log transformed the income and expenses change measures (ln + 1), which
significantly increases the fit of our full model across 3, 12, and 36 months (R2 = .039, .201, .136
respectively), compared to using raw values (R2 = .009, .111, .072 respectively).
For each time period, we ran a regression with income change and expense change to
predict slack change. For the 3 months’ time frame, the coefficient on Income Change (β = 0.494,
SE = .228, t(115) = 2.17, p = .032) is 5.31 times the size of the coefficient on Expense Change (β
= -0.093, SE = .177, t(115) = -0.52, p = .60). We then compared the Full Model (R2 = .039) that
allowed the income and expense parameters to vary to the Constrained Model where income and
expenses have equal and opposite signs (R2 = .017), however this difference was not significant
(F(1, 116) = 2.69, p = .10).
For the 36 months’ time frame, the coefficient on Income Change (β = 0.977, SE = .237,
t(116) = 4.13, p < .001) is 2.82 times the size of the coefficient on Expense Change (β = -0.346,
SE = .261, t(116) = -1.32, p = .19). Further, the full model (R2 = .136) explains a significantly
greater amount of variance than the constrained model (R2 = .084; F(1, 116) = 6.86, p = .010),
providing evidence for expense neglect over 36 months.
Response Scale Condition: Participants on average expected their slack to grow in 3
months (M = 0.71, SD = 1.87; one-sample t(124) = 4.26, p < .001), 12 months (M = 1.44, SD =
2.57; one-sample t(124) = 6.26, p < .001), and 36 months (M = 2.57, SD = 3.13; one-sample
EXPENSE NEGLECT – WEB APPENDIX
5
t(124) = 9.18, p < .001). Participants expected their income to grow in 3 months (M = 0.81, SD =
1.57; one-sample t(124) = 5.74, p < .001), 12 months (M = 1.70, SD = 1.63; one-sample t(124) =
11.62, p < .001), and 36 months (M = 2.69, SD = 1.78; one-sample t(124) = 16.93, p < .001).
Consistent with the Modulus condition, participants did not expect their expenses to grow
significantly in 3 months (M = 0.08, SD = 1.37; one-sample t(124) = 0.65, p = .52), but did
expect their expenses to grow in 12 months (M = 0.68, SD = 1.66; one-sample t(124) = 4.58, p
< .001) and 36 months (M = 1.59, SD = 2.03; one-sample t(124) = 8.74, p < .001).
For each time period, we ran a regression with income change and expense change to
predict slack change. For the 3 months’ time frame, The coefficient on Income Change (β =
0.670, SE = .094, t(122) = 7.10, p < .001) is 2.10 times the size of the coefficient on Expense
Change (β = -0.319, SE = .108, t(122) = -2.94, p = .004). Further, the full model (R2 = .294)
explains a significantly greater amount of variance than the constrained model (R2 = .248; F(1,
122) =7.88, p = .006), providing evidence for expense neglect over 3 months.
For the 36 months’ time frame, the coefficient on Income Change (β = 1.105, SE = .129,
t(122) = 8.54, p < .001) is 2.72 times the size of the coefficient on Expense Change (β = -0.406,
SE = .113, t(122) = -3.60, p < .001). Further, the full model (R2 = .382) explains a significantly
greater amount of variance than the constrained model (R2 = .271; F(1, 122) = 21.73, p < .001),
providing evidence for expense neglect over 36 months.
STUDY 5: MEASURING INCOME AND EXPENSES IN DOLLARS DOES NOT
ELIMINATE EXPENSE NEGLECT
Method
EXPENSE NEGLECT – WEB APPENDIX
6
We recruited 121 individuals (27.0% female; mean age = 29.2; median household income
= $40 to $50k) via Mechanical Turk to participant in a study about how people make estimations.
Participants were asked to answer four separate focal questions asking participants to estimate
their income and expenses next month and in 24 months in dollars. These four questions were
embedded within a series of 15 total estimates such as “How many miles do you think you will
drive in a car next month?” This procedure was implemented in order to mask the purpose of the
study, so that participants did not reduce the task to a simple algebra problem. After making
these estimates, participants then evaluated their financial slack using the three-item measure.
Results
Participants on average expected their slack (M = 1.76, SD = 2.51; one-sample t(121) =
7.73, p < .001), income (Median = +$1,000; Wilcoxon signed rank test Z = 8.61, p < .001), and
expenses (Median = +$400; Wilcoxon signed rank test Z = 7.88, p < .001) to grow over the next
12 months. Further, participants reported having greater current income (Median = $1,850) than
expenses (Median = $1,000; Wilcoxon signed rank test Z = 6.50, p < .001).
STUDY 6: UNCERTAINTY DOES NOT EXPLAIN EXPENSE NEGLECT
Method
We were concerned about possible ceiling effects in reports of confidence in estimates of
income and expense growth. As a result, at the beginning of the survey, participants were given a
reference question meant to extend the range of the confidence scales. All participants indicated
the likelihood that they would be able to find time to make a 10-minute call in the following
week on a scale ranging from 1 = “Very unlikely” to 11 = “Extremely likely.” Participants were
EXPENSE NEGLECT – WEB APPENDIX
7
then asked, “How confident are you in your response?” on a scale ranging from 1 = “Not at all
confident” to 11 = “Extremely confident.” Participants reported being highly likely to make the
call (M = 10.26, SD = 1.44) and highly confident in their response (M = 10.34, SD = 1.23).
STUDY 9: SPENDTHRIFTS ARE MORE PRONE TO EXPENSE NEGLECT THAN
TIGHTWADS
Pre-Test
We initially recruited 151 participants via Mechanical Turk to examine how individual
differences in the T-S scale related to expense neglect. Participants were asked to evaluate their
income, expenses, and slack next month and in two years. Participants were given the T-S scale.
We examine if there was a significant expense change by T-S interaction in predicting slack
change. An interaction would show that individuals who vary on the T-S scale would also weight
expenses differently when forecasting future slack. Specifically, we tested to see if tightwads are
more likely to properly weigh expenses than spendthrifts. The pre-test found results that the
expense change by T-S interaction was directionally, but not significantly significant with our
predictions (β = 0.018, SE = .014, t(146) = 1.26, p = .21). As a result, we ran a power analysis
based on the pre-test results (Cohen’s f2 = 0.0137; α = .05; 1- β = .80) to compute the sample size
necessary to achieve sufficient power. The power analysis results estimated a sample size of 942
participants. We chose to recruit a new sample of 1,000 participants.
EXPENSE NEGLECT – WEB APPENDIX
8
WEB APPENDIX B
MANIPULATING ATTENTION
Method
We conducted two versions of the same study design. In the online version, we recruited
242 participants (40.0% female; mean age = 31.4; median household income = $40 to $50k) via
Mechanical Turk to participate in a study about personal finances. In the lab version, we
recruited 156 participants (71.1% female; mean age = 21.2) from a northeastern university.
In each study, participants were allocated in one of four conditions. In all conditions
participants evaluated their spare money in the present month and in 24 months from now on an
11-point scale ranging 1 = “Very little available money” to 11 = “Lots of available money”.
Participants also evaluated how much their income and expenses will increase and decrease in 24
months on a scale ranging from -5 = “I expect that my income [expenses] will greatly decrease”
to +5 = “I expect that my income [expenses] will greatly increase.”
In each condition, we manipulated participants’ attention to the various inputs into slack
and their attention to how these inputs may change over time. Participants in the Control
condition evaluated their slack first and then estimated their income and expense growth after
evaluating their slack. Participants in Category condition evaluated their income and expense
growth prior to estimating their slack. Participants in the Subcategory condition first evaluated
the percentage of their total income that is made up of various subcategories (salary and wages;
social security; alimony and child support; returns on investments; and other), and the percentage
of total expenses that is made up from subcategories (rent and mortgage payments; food and
groceries; utilities; transportation and automobile expenses; entertainment; savings
commitments; and other). These participants then evaluated their income and expense growth
EXPENSE NEGLECT – WEB APPENDIX
9
followed by their financial slack. Finally, participants in the Forecasting condition evaluated the
components of their income and expense as with the Subcategory condition, and also estimated
how each of these categories would grow or shrink in 24 months.
MTurk Study Results
Predictions of Slack Growth: Collapsing across all conditions, participants expected their
slack (M = 2.32, SD = 2.76), income (M = 2.12, SD = 1.92), and expenses (M = 1.40, SD = 2.07)
to grow in the future (one-sample ts > 10.5, ps < .001) relative to next month.
Next we tested to see if greater attention to inputs (both income and expenses) would lead
to participants attending to expenses and weighting it more. We ran the following regression:
(1) Slack Change = β0(Intercept) + β1(Income Change) + β2(Expense Change) + β3-5(Condition)
+ β6-8(Income Change * Condition) + β9-11(ExpenseChange * Condition)
A significant condition by income or condition by expenses interaction would suggest
that our manipulation of attention is affecting the degree to which participants’ evaluation of
income or expenses affected slack growth. An analysis of covariance of the above model showed
a marginally significant expense change by condition interaction (F(3,230) = 2.47, p = .062),
whereas the income change by condition interaction was not significant (F(3,230) = 0.83, p
= .48). Further analysis shows that the increased attention to inputs did not eliminate expense
neglect in any condition. Table 1 presents the results from regressions ran for each condition
separately. Note that in each of the conditions, the coefficient on income change is in the
predicted direction and significant. However, the coefficient on expense change is not significant
EXPENSE NEGLECT – WEB APPENDIX
10
in any condition except for the control condition, where it is in the opposite direction of what is
predicted.
Table 1: MTurk Study
Regression Coefficients
Model Comparison
Condition
N
Intercept
Income
Change
Expense
Change
F
p
Control
57
-6.58***
0.92***
0.26*
50.99
< .001
(1.32)
(0.14)
(0.11)
2.66
.11
26.37
< .001
9.52
.003
Category
Subcategory
Forecasting
59
66
60
-1.42
0.65**
-0.25
(1.93)
(0.20)
(0.20)
-3.95**
0.81***
-0.01
(1.18)
(0.12)
(0.12)
-3.50*
0.86***
-0.21
(1.69)
(0.19)
(0.19)
Notes: Regression output for individual conditions in the MTurk version of the study. Parenthesis
indicate standard errors. * p < .05, ** p < .01, *** p < .001
Lab Study Results
Predictions of Slack Growth: Collapsing across all conditions, participants expected their
slack (M = 2.09, SD = 3.52), income (M = 3.03, SD = 2.01), and expenses (M = 2.38, SD = 2.11)
to grow in the future (one-sample ts > 10.5, ps < .001) relative to next month.
Next we tested to see if greater attention to inputs (both income and expenses) would lead
to participants attending to expenses and weighting it more. To do so, we ran the same model as
used above.
EXPENSE NEGLECT – WEB APPENDIX
11
An analysis of covariance showed a significant expense change by condition interaction
(F(3,144) = 3.56, p = .030), whereas the income change by condition interaction was not
significant (F(3,144) = 1.88, p = .14).
Table 2 presents the results from regressions ran for each condition separately. While the
weight placed on expenses increases in the Subcategory and the Forecasting condition relative to
the Control and Category conditions, no reliable patterns emerged.
Table 2: Lab Study
Regression Coefficients
Condition
N
Intercept
Income
Change
Expense
Change
Control
40
-1.46
(3.46)
0.37
(0.35)
Category
38
-5.99
(3.31)
Subcategory
38
Forecasting
40
Model Comparison
F
p
0.03
(0.30)
1.09
.30
0.42
(0.30)
0.47
(0.26)
5.31
.027
0.70
(2.17)
0.66**
(0.20)
-0.51*
(0.23)
0.32
.58
-7.68
(3.24)
1.42***
(0.37)
-0.42
(0.39)
7.82
.008
Notes: Regression output for individual conditions in the MTurk version of the study.
Parenthesis indicate standard errors. * p < .05, ** p < .01, *** p < .001
EXPENSE NEGLECT – WEB APPENDIX
12
WEB APPENDIX C
META-ANALYSIS FULL DETAILS
The tables and charts on the following pages present the data used in for each metaanalysis reported in the main manuscript. Table 3 presents a summary of the studies used in the
meta-analysis. Tables 4 & 5 presents the data used to determine the effect size of the model
comparison across the entire set of studies. In each study, we compare the full model in which
income and expenses can vary to the restricted model in which income and expenses have equal
and opposite signs. Each study contributed three partial correlations into the meta-analysis,
which were all converted into Fisher’s Z:

The semi-partial correlation of the model comparison:
Full Model:
Slack Changei = β0 + β1Income Changei + β2Expense Changei + i
Constrained Model: Slack Changei = β0 + β1(Income Changei - Expense Changei) + i
2
2
Semi-partial R = √𝑅𝐹𝑢𝑙𝑙
− 𝑅𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑒𝑑

The partial r between Income Change and Slack Change from the full model.

The partial r between Expense Change and Slack Change from the full model.
We computed Fisher’s Z for each partial correlation from each study and calculated a
weighted mean Z, using inverse variance weights to assign more weight to studies with larger
samples. Table 4 examines the model comparison statistics for slack change, specifically whether
income change and expense change are weighted equally when predicting slack change. Table 5
EXPENSE NEGLECT – WEB APPENDIX
13
examines the individual effect size for the weight on income change and the weight on expense
change when predicting slack change. Figure 1 presents a summary of the results from Table 5.
Tables 6 – 9 examine the “static” relationship between income, expenses and slack.
Specifically, they examine the weight placed on income and expenses when predicting slack at a
single point in time (they do not examine growth in income, expenses, and slack over time).
These include only the subset of studies that measure income, expenses, and slack in the
immediate future (e.g., next month) and the distant future (e.g., 24 months from now), as noted
in the Appendix of the main manuscript under the “Income/Expense Measures” column.
Each study contributed six partial correlations into the meta-analysis, which included
three partial correlations for the immediate future time period and three partial correlations for
the distant future time period, each of which were all converted into Fisher’s Z:

The semi-partial correlation of the model comparison:
Full Model:
Slackit = β0 + β1Incomeit + β2Expensesit + i
Constrained Model: Slackit = β0 + β1(Incomeit -Expensesit) + i
2
2
Semi-partial R = √𝑅𝐹𝑢𝑙𝑙
− 𝑅𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑒𝑑

The partial r between Income and Slack from the full model.

The partial r between Expense and Slack from the full model.
Table 6 and 7 represent the data for the immediate future, where slack, income, and
expenses are evaluated in 1 month from the present (i.e., next month). Table 6 examines the
model comparison statistics for the immediate future, specifically whether income and expenses
are weighted equally when predicting slack in the immediate future. Table 7 examines the
EXPENSE NEGLECT – WEB APPENDIX
14
individual effect size for the weight on income and the weight on expenses when predicting slack
in the immediate future.
Table 8 and 9 represent data for the distant future, where slack, income, and expenses are
measured in 9 – 24 months from the present. Table 8 examines the model comparison statistics
for the distant future, specifically whether income and expenses are weighted equally when
predicting slack in the distant future. Table 9 examines the individual effect size for the weight
on income and the weight on expenses when predicting slack in the distant future.
EXPENSE NEGLECT – WEB APPENDIX
15
EXPENSE NEGLECT – WEB APPENDIX
Table 4: Model Comparison for Slack Change
ManuStudy script
#
Study Name
Time
Periods
(Mo. From
Present)
N
R Full
Model
2
Semi-partial r R2
Model
Constrained
Comparison
Model
Fisher's Z
Weight x
Effect Size
A
1
Initial Demonstration I
0, 12
580
0.11
0.05
0.23
0.24
137.84
B
-
Initial Demonstration II
1, 9
135
0.27
0.15
0.35
0.36
48.03
C-1
4
Modulus Study I - Modulus Scale
0, 12
119
0.20
0.12
0.28
0.29
33.54
C-2
4
Modulus Study I - Likert Scale
0, 12
125
0.37
0.34
0.17
0.17
20.67
D-1
-
Modulus Study II - Modulus Scale
0, 12
107
0.07
0.00
0.25
0.25
26.19
D-2
-
Modulus Study II - Likert Scale
0, 12
101
0.10
0.06
0.18
0.18
17.85
E
WA-B
Category Attention Study I
0, 24
242
0.28
0.10
0.42
0.45
107.50
F
WA-B
Category Attention Study II
0, 24
156
0.12
0.07
0.24
0.24
37.05
G
-
Past vs Present
0, 24
153
0.24
0.22
0.14
0.14
21.56
H
-
Inflation Study I
0, 12
176
0.13
0.12
0.05
0.05
8.48
I
-
Exceptional Expenses I
0, 12
201
0.04
0.04
0.04
0.04
8.41
J
-
Expense vs. Expenditure
1, 9
485
0.09
0.08
0.07
0.07
34.63
K
-
Exceptional Expenses II
0, 12
184
0.06
0.05
0.09
0.09
16.92
L
-
Exceptional Expenses III
0, 12
157
0.03
0.03
0.01
0.01
1.33
M
-
Confidence Study I
1, 24
306
0.16
0.11
0.21
0.22
65.28
N
6
Confidence Study II
1, 24
301
0.27
0.20
0.28
0.29
86.37
O
3
Constraint I
1, 24
74
0.36
0.30
0.25
0.25
18.06
P
-
Dollar I
1, 24
77
0.06
0.04
0.14
0.14
10.39
Q
-
Constraint II
1, 24
152
0.17
0.17
0.01
0.01
0.76
R
2
Constraint vs Slack
1, 24
302
0.19
0.15
0.19
0.19
58.23
S
5
Dollar II
1, 24
121
0.09
0.04
0.24
0.25
29.09
T
-
Inflation Study II
1, 24
154
0.06
0.06
0.05
0.05
7.39
U
-
Inflation Study III
1, 24
150
0.27
0.18
0.30
0.31
45.71
V
WA-A
Tightwad I
1, 24
151
0.29
0.19
0.31
0.32
47.31
W
9
Tightwad II
1, 24
1000
0.23
0.18
0.22
0.23
226.60
X
8
Optimism Study
1, 24
1005
0.27
0.20
0.27
0.28
277.62
Y
7
Flexibility Study
1, 24
500
0.08
0.02
0.25
0.25
Total
7214
126.41
1519.21
16
EXPENSE NEGLECT – WEB APPENDIX
Table 5: Partial Effect of Income Change and Expense Change on Slack Change
ManuStudy script
#
A
1
N
Income Δ Parial r w/
Slack
Change
Fisher's Z Income Δ
0, 12
580
0.33
0.34
196.45
-0.03
-0.03
Study Name
Time
Periods
(Mo. From
Present)
Initial Demonstration I
Expense Δ Weight x
Weight x
Parial r w/ Fisher's Z Effect Size Effect Size Slack
Expense Δ
Income Δ
Expense Δ
Change
-14.47
B
-
Initial Demonstration II
1, 9
135
0.52
0.58
76.42
-0.06
-0.06
-7.68
C-1
4
Modulus Study I - Modulus Scale
0, 12
119
0.44
0.48
55.23
-0.11
-0.11
-13.19
C-2
4
Modulus Study I - Likert Scale
0, 12
125
0.58
0.66
80.38
-0.42
-0.45
-54.54
D-1
-
Modulus Study II - Modulus Scale
0, 12
107
0.21
0.21
22.11
0.11
0.11
11.59
D-2
-
Modulus Study II - Likert Scale
0, 12
101
0.30
0.31
30.14
-0.05
-0.05
-4.93
E
WA-B
Category Attention Study I
0, 24
242
0.52
0.58
137.78
-0.02
-0.02
-4.74
F
WA-B
Category Attention Study II
0, 24
156
0.35
0.37
56.14
-0.09
-0.09
-13.87
G
-
Past vs Present
0, 24
153
0.49
0.54
80.53
-0.22
-0.23
-33.91
H
-
Inflation Study I
0, 12
176
0.32
0.33
57.29
-0.23
-0.23
-40.52
I
-
Exceptional Expenses I
0, 12
201
0.13
0.13
25.44
-0.17
-0.17
-33.75
J
-
Expense vs. Expenditure
1, 9
485
0.26
0.27
129.29
-0.16
-0.16
-79.10
K
-
Exceptional Expenses II
0, 12
184
0.23
0.23
42.08
-0.13
-0.13
-23.97
L
-
Exceptional Expenses III
0, 12
157
0.14
0.14
21.32
-0.13
-0.13
-19.42
M
-
Confidence Study I
1, 24
306
0.40
0.42
127.32
-0.18
-0.18
-55.52
N
6
Confidence Study II
1, 24
301
0.52
0.58
173.35
-0.17
-0.17
-51.04
O
3
Constraint I
1, 24
74
0.60
0.69
48.70
-0.35
-0.36
-25.56
P
-
Dollar I
1, 24
77
0.15
0.15
10.82
-0.24
-0.24
-17.85
Q
-
Constraint II
1, 24
152
0.33
0.34
50.91
-0.33
-0.34
-50.41
R
2
Constraint vs Slack
1, 24
302
0.42
0.45
133.45
-0.19
-0.20
-59.05
S
5
Dollar II
1, 24
121
0.24
0.24
28.78
0.06
0.06
6.96
T
-
Inflation Study II
1, 24
154
0.23
0.24
35.63
-0.16
-0.16
-24.16
U
-
Inflation Study III
1, 24
150
0.52
0.57
83.88
-0.16
-0.16
-24.07
V
WA-A
Tightwad I
1, 24
151
0.54
0.60
88.42
-0.13
-0.13
-18.81
W
9
Tightwad II
1, 24
1000
0.46
0.50
499.95
-0.20
-0.20
-201.70
X
8
Optimism Study
1, 24
1005
0.52
0.57
576.12
-0.20
-0.21
-206.27
Y
7
Flexibility Study
1, 24
500
0.26
0.27
133.86
0.03
0.03
13.42
Total
7214
3001.78
-1046.56
17
EXPENSE NEGLECT – WEB APPENDIX
Income Change (Fisher's Z)
Expense Change (Fisher's Z)
1.00
0.80
0.60
0.40
Fisher's Z
0.20
0.00
A
B C-1C-2D-1D-2 E
F G H
I
J
K
L M N O P Q R S
T
U V W X
Y
-0.20
-0.40
-0.60
-0.80
-1.00
Figure 1: Standardized effect sizes for the weight of income change and expense change for each individual study, as reported in
Table 5.
18
EXPENSE NEGLECT – WEB APPENDIX
Table 6: Model Comparison for Slack in the Immediate Future
Time
Manu- Period
Study script
(Mo.
#
From
Now)
B
Study Name
N
R2 Full
Model
2
Semi-partial
R
Constrained r of Model
Comparison
Model
Fisher's Z
Weight x
Effect Size
-
1
Initial Demonstration II
135
0.42
0.22
0.45
0.48
63.48
J
-
1
Expense vs. Expenditure
485
0.27
0.23
0.21
0.21
103.60
N
6
1
Confidence Study II
301
0.27
0.21
0.24
0.25
74.23
O
3
1
Constraint I
74
0.43
0.40
0.16
0.16
11.47
P
-
1
Dollar I
77
0.05
0.03
0.12
0.12
8.59
Q
-
1
Constraint II
152
0.30
0.30
0.02
0.02
2.94
R
2
1
Constraint vs Slack
302
0.31
0.27
0.18
0.18
54.22
S
5
1
Dollar II
121
0.11
0.08
0.19
0.19
22.12
T
-
1
Inflation Study II
154
0.31
0.26
0.22
0.23
34.15
U
-
1
Inflation Study III
150
0.27
0.23
0.19
0.19
27.60
V
WA-A
1
Tightwad I
151
0.44
0.33
0.33
0.34
50.66
W
9
1
Tightwad II
1000
0.28
0.26
0.15
0.15
149.50
X
8
1
Optimism Study
1005
0.32
0.29
0.17
0.17
174.50
Y
7
1
Flexibility Study
500
0.22
0.19
0.19
0.19
Total
4607
95.46
872.53
19
EXPENSE NEGLECT – WEB APPENDIX
Table 7: Partial Effect of Income and Expenses on Slack in the Immediate Future
Time
Manu- Period
Study script (Mo.
#
From
Now)
Study Name
N
Income Parial r w/
Slack
Fisher's Z Income
Weight x
Effect Size
Expenses
Fisher's Z - Parial r
Expenses
w/ Slack
Initial Demonstration II
135
0.64
0.76
100.75
-0.18
-0.18
-23.89
Expense vs. Expenditure 485
0.47
0.52
248.33
-0.23
-0.23
-110.34
Weight x
Effect Size
B
-
1
J
-
1
N
6
1
Confidence Study II
301
0.51
0.56
168.10
-0.25
-0.26
-76.75
O
3
1
Constraint I
74
0.63
0.75
52.99
-0.48
-0.53
-37.41
P
-
1
Dollar I
77
0.20
0.20
14.62
-0.20
-0.21
-15.23
Q
-
1
Constraint II
152
0.47
0.51
76.00
-0.45
-0.49
-72.78
R
2
1
Constraint vs Slack
302
0.53
0.59
176.87
-0.36
-0.37
-111.32
S
5
1
Dollar II
121
0.31
0.32
37.43
-0.15
-0.15
-17.35
T
-
1
Inflation Study II
154
0.55
0.62
93.38
-0.35
-0.36
-55.01
U
-
1
Inflation Study III
150
0.50
0.55
80.55
-0.34
-0.35
-51.22
V
WA-A
1
Tightwad I
151
0.54
0.60
88.58
-0.36
-0.37
-55.44
W
9
1
Tightwad II
1000
0.50
0.55
545.00
-0.33
-0.34
-342.92
X
8
1
Optimism Study
1005
0.54
0.60
599.72
-0.36
-0.37
-375.34
Y
7
1
Flexibility Study
500
0.46
0.49
244.65
-0.24
-0.24
-121.13
Total
4607
2526.98
-1466.12
20
EXPENSE NEGLECT – WEB APPENDIX
Table 8: Model Comparison for Slack in the Distant Future
ManuStudy script
#
Time
Period
(Mo.
From
Now)
Study Name
N
R2 Full
Model
2
Semi-partial
R
Constrained r of Model
Comparison
Model
Fisher's Z
Weight x
Effect Size
B
-
9
Initial Demonstration II
135
0.39
0.32
0.27
0.28
35.76
J
-
9
Expense vs. Expenditure
485
0.20
0.14
0.25
0.25
121.83
N
6
24
Confidence Study II
301
0.28
0.23
0.21
0.22
63.69
O
3
24
Constraint I
74
0.16
0.11
0.21
0.21
14.58
P
-
24
Dollar I
77
0.28
0.21
0.27
0.27
19.41
Q
-
24
Constraint II
152
0.34
0.22
0.34
0.35
51.77
R
2
24
Constraint vs Slack
302
0.28
0.28
0.04
0.04
11.52
S
5
24
Dollar II
121
0.44
0.30
0.38
0.40
45.44
T
-
24
Inflation Study II
154
0.44
0.33
0.32
0.33
48.95
U
-
24
Inflation Study III
150
0.32
0.21
0.34
0.35
50.48
V
WA-A
24
Tightwad I
151
0.07
0.02
0.22
0.23
32.83
W
9
24
Tightwad II
1000
0.04
0.01
0.17
0.17
168.19
X
8
24
Optimism Study
1005
0.45
0.37
0.27
0.28
281.49
Y
7
24
Flexibility Study
500
0.53
0.48
0.24
0.24
119.66
Total
4607
1065.61
21
EXPENSE NEGLECT – WEB APPENDIX
Table 9: Partial Effect of Income and Expenses on Slack in the Distant Future
Time
Manu- Period
Study script
(Mo.
#
From
Now)
Study Name
N
Income Parial r w/
Slack
Fisher's Z Income
Weight x
Effect Size
Expenses Fisher's Z Parial r w/
Expenses
Slack
Weight x
Effect Size
B
-
9
Initial Demonstration II
135
0.61
0.64
84.69
-0.16
-0.16
-21.44
J
-
9
Expense vs. Expenditure
485
0.45
0.57
276.48
-0.27
-0.28
-135.01
N
6
24
Confidence Study II
301
0.52
0.58
174.21
-0.20
-0.20
-60.41
O
3
24
Constraint I
74
0.40
0.88
62.56
-0.56
-0.63
-44.42
P
-
24
Dollar I
77
0.53
0.15
10.73
0.06
0.06
4.52
Q
-
24
Constraint II
152
0.58
0.50
73.91
-0.29
-0.30
-44.97
R
2
24
Constraint vs Slack
302
0.46
0.42
126.32
-0.18
-0.18
-52.87
S
5
24
Dollar II
121
0.66
0.27
32.03
-0.13
-0.13
-15.19
T
-
24
Inflation Study II
154
0.66
0.66
99.80
-0.19
-0.19
-28.57
U
-
24
Inflation Study III
150
0.57
0.79
116.02
-0.32
-0.33
-48.59
V
WA-A
24
Tightwad I
151
0.27
0.80
118.13
-0.20
-0.20
-30.16
W
9
24
Tightwad II
1000
0.14
0.72
713.17
-0.27
-0.28
-278.19
X
8
24
Optimism Study
1005
0.66
0.79
792.63
-0.33
-0.35
-345.76
Y
7
24
Flexibility Study
500
0.71
0.48
237.79
-0.13
-0.13
-62.96
Total
4607
2918.47
-1164.02
22