Bias in retrospective reports of cohabitation among new parents Julien O. Teitler Nancy E. Reichman Heather Koball May 2004 This research was supported by the National Institute of Child Health and Human Development (R01-HD-35301). Comparison of Contemporaneous and Retrospective Reports of Cohabitation Abstract: Most estimates of cohabitation use retrospective survey reports. We compare contemporaneous and retrospective reports of cohabitation and assess the implications for studying relationship change. We find that (1) many mothers revise their reports of whether or not they cohabited at the time of the birth of their child; (2) changes in reports are not simply due to random measurement error, but are a function of individual characteristics and the couple’s subsequent relationship status; and (3) estimates of cohabitation rates at birth and some analyses of union formation and dissolution are sensitive to whether contemporaneous or retrospective reports are used. 1 Non-marital cohabitation has increased substantially over the past several decades, as evidenced by analyses of the National Survey of Family Growth (NSFG), the National Survey of Families and Households (NSFH), and the Current Population Surveys (CPS) (Bumpass and Sweet, 1989a; Bumpass and Lu, 2000; Casper and Cohen, 2000). These nationally representative surveys include data on living arrangements and can be used to estimate cohabitation rates of unmarried parents at the time their children are born. However, the literature on recall bias in surveys suggests that retrospective reports may be biased. We use data from the Fragile Families and Child Wellbeing – a recent national longitudinal survey of new parents – to assess the degree to which retrospective reports of cohabitation status with a partner one year earlier reflect individual characteristics and current aspects of the relationship with that partner. We also assess the extent of retrospective reporting bias and it affects analyses of union formation and dissolution. We situate our findings within the literature on autobiographical memory. This is the first study to assess the implications of using retrospective reports of cohabitation. BACKGROUND During the 1980s, only 25% of all non-marital births were to cohabiting parents (Bumpass and Sweet, 1989a). Recent estimates indicate that 39% of unmarried parents are cohabiting at the time their children are born, with substantial variation by race/ethnicity (Bumpass and Lu 2000). The rates among unmarried non-Hispanic whites and Hispanics are 50 and 53%, respectively, while the rate among non-Hispanic blacks is less than half that (22%) (Bumpass and Lu 2000). The increase in cohabitation rates since then has been greatest for non- 2 Hispanic whites (from 33% to 50%) (Bumpass and Lu, 2000). Casper and Cohen (2000) show similar increases for the period 1977 to 1997. Cohabitation is a difficult concept to measure for a variety of reasons. Direct and indirect measures of cohabitation can yield different estimates (Baughman, Dickert-Conlin, and Houser, 2002; Fields and O'Connell, 2003; Teitler and Reichman, 2001), as can contemporaneous and retrospective reports (Teitler and Reichman, 2001). A growing body of research suggests that it may be difficult to ascertain cohabitation status because of conceptual ambiguities. Non-marital relationships do not fall clearly into rigid one-or-the-other categories, but occur along a continuum (Edin and Lein, 1997; Ross, 1995; Seltzer, 2000). It is not always clear to couples themselves when they began cohabiting (Manning and Smock, 2003) and many cycle in and out of co-residence (Binstock and Thornton, 2003). Parents sometimes disagree with one another about whether or not they are presently living together (Nock, 1995; Seltzer, 2000) and whether or not they live with their child (Lin et al., forthcoming). The meaning of cohabitation also may vary depending on the race, ethnicity, age, educational level, and income of respondents (Manning, 1993; Smock, 2000). The CPS collects current household roster information and asks about cohabitation, but as Casper and Cohen (2000) point out, estimates are very sensitive to how cohabitation status is inferred from the rosters since these ask about cohabitation in relation to the head of the household. The NSFG and NSFH collect relationship histories (up to 5 years prior to the interview in the NSFG and even farther back than that in the NSFH) that make it possible to estimate cohabitation rates of unmarried parents at the time of a birth. However, a large body of research on autobiographical memory and the validity of retrospective reports makes it clear 3 that recall, even of important life events, is often imperfect and suggests that the estimates of cohabitation based on retrospective reports might be biased. The extent of bias in retrospective reports of cohabitation is not known. It is well known, however, that recall is affected by a number factors, including the importance of the topic or event to the respondent, how far back in time the respondent is asked to remember, the ambiguity or complexity of the concept being asked about, the social desirability of what is being asked about, and the attention and motivation of the respondent (see Groves 1989, Schaeffer 1994, and Shum and Ripps 1999). When memory retrieval is difficult, the respondent may rely on variety of reconstructive strategies, some of which may give rise to reporting anomalies such as the seam effect in panel data (Rips, Conrad and Fricker 2003). Since relationships and living arrangements are likely to be salient to individuals, and recalling them does not seem to be a highly complex task, one might expect retrospective reports of cohabitation status to yield reasonably accurate data. Also, since recall of past events is facilitated when important events can be used as markers (Loftus and Marburger, 1983), asking about cohabitation status at the time of a child’s birth, a landmark event, may further reduce potential recall bias. Finally, although cohabitation is an elusive concept that is subject to individual interpretation (as discussed earlier), it is unclear whether this complexity makes it more difficult to ascertain in retrospect; similarly, it is not clear whether perceived social desirability would affect contemporaneous and retrospective reports of cohabitation differentially. Whereas the literature cited above does not necessarily indicate that retrospective reports of cohabitation would be systematically biased, psychological studies of memory reconstruction suggest otherwise. Ross (1989) claimed that when respondents have trouble 4 recalling events, they tend to assume consistency in states unless they have strong reasons to assume changes would have occurred, in which case they may exaggerate reports of change. We might expect, then, that mothers who do not recall exactly what their living arrangement was at the time of their child’s birth would assume it was similar to the current status, unless their perception of the father had changed substantially since that time. According to Hasher and Griffen (1978), reports of past circumstances may be related to feelings about current circumstances, and according to Bower (1987) and Isen (1987), they can be influenced by current mood. Particularly relevant is a study by Holmberg and Holmes (1994) that found that memory reconstruction of past relationships is related to current affect. In this paper, we examine the extent, determinants, and consequences of bias in retrospective reports of cohabitation in the context of this literature on autobiographical memory. Lillard and Waite (1988) compared retrospective and contemporaneous reports of marriage in the Panel Survey of Income Dynamics, and found non-trivial levels of disagreement (both over and under-estimates) in both the number of past marriages and when those marriages occurred. Peters (1988) also found systematic discrepancies between panel and retrospective reports of marriage. Given the ambiguities surrounding the concept of cohabitation, we expect that retrospective reports of cohabitation are subject to even more bias than are those of marriage. Based on the extensive research on cohabitation and recall in surveys spanning multiple decades, we hypothesize that mothers whose relationships have deteriorated will be more likely to retrospectively report that they were not living together at birth and that those whose relationships have remained stable or improved will be more likely to report having cohabited at birth. If recall of past relationships does indeed operate this way, retrospective reports of 5 cohabitation might be biased and rates of union formation and dissolution could be underestimated. DATA We use panel data from the Fragile Families and Child Wellbeing study, which follows a cohort of 4898 sets of new parents and their children in 20 US cities. The study was designed to take a longitudinal look at the nature, determinants, and trajectories of relationships of new (mostly unwed) parents. Mothers were interviewed in-person in the hospital at the time of the birth, again by telephone when the child was 12-18 months old, and periodically every few years thereafter. The data are representative of births in large US cities (see Reichman et al. 2001 for a description of the Fragile Families research design). Non-marital births were oversampled (by a factor of approximately three) by design, yielding a sample of 3712 sets of unmarried parents. Ninety percent of eligible unmarried mothers completed baseline interviews, and 87 percent of those mothers completed interviews at the first follow-up. The baseline interviews elicited information about cohabitation a number of different ways. Each mother was asked about her relationship status with the biological father of her new baby, and those who reported that they were “romantically involved now” or involved “in an on again, off again relationship” were asked whether they were currently living with the father. The mother was asked for information with which the interviewer could construct a household roster that included the mother’s relationship to each person. We use some of these questions to construct four measures of baseline cohabitation. One is the mother’s direct report of whether she was living with the father. Another is whether she listed the father on the household roster. In addition, we combined these two measures to create a conservative measure of cohabitation, 6 whether the mother reported that she lives with the father and listed him on the household roster, and a liberal measure, whether she reported that she lives with the father or listed him as a member of her household. Estimates using the four different measures, shown in Table 1, indicate that cohabitation rates are fairly insensitive to the baseline measure used. Both the direct question and the household roster result in virtually identical estimates (48% and 47%, respectively). The range is only slightly larger when we compare the most and least restrictive measures (45 to 50%). Therefore, in all remaining analyses we use the baseline question that is most comparable to the follow-up question about baseline cohabitation (the direct question about whether the parents were living together). In general, Fragile Families estimates of cohabitation at birth are higher than those using national data sets, even when taking into consideration the age distribution and urban nature of the sample (McLanahan et al., 2001). The discrepancy may be due to differences in measurement, or it may reflect period effects (Fragile Families estimates are based on more recent births). [Table 1 here] Next we compare retrospective and contemporaneous survey reports of cohabitation among mothers giving birth out of wedlock. At the one year follow-up interview, in addition to being asked about current relationship status and living arrangements, mothers were also asked whether they lived with the father when their child was born. However, the response categories were expanded to distinguish between living together “all or most of the time,” and “some of the time.”1 In most analyses that follow, we address this inconsistency in response categories 1 The exact questions are as follows: At baseline, if the mother indicated that she was in a romantic relationship or an “on again, off again” relationship with the father, she was asked “Are you and [BABY’S FATHER] living 7 by excluding the respondents who at the one-year follow-up reported living with the father some of the time at baseline. We test the sensitivity of our findings to this assumption by conducting supplemental analyses that include mothers who reported that they were living together some of the time as cohabiting, and others in which this group is considered not cohabiting. We are interested in whether changes in reports are systematically related to the mother’s own characteristics, attributes of the baby’s father, and the quality and duration of their relationship since the baseline interview. Based on past literature revealing subgroup differences in reports of cohabitation, we assess whether revisions of reports vary by race/ethnicity, nativity, and education. Specifically, we consider whether the mother is black non-Hispanic, Hispanic, or other non-white non-Hispanic (compared to white non-Hispanic); whether the mother was born in the US (versus foreign born); and whether the mother had less than a high school education or had some college at the time of the birth (compared to being a high school graduate). We also include a measure of the number of months between the mother’s baseline and follow-up interviews; we expect that increased time elapsed will be associated with less accurate recall due to the direct effects of time on actual memory, or through changes in the parents’ relationship and life circumstances that may lead the mother to reconstruct her past. We also consider whether the propensity to revise reports of baseline cohabitation varies by characteristics of the baby’s father. We consider whether the father was unemployed at the time of the follow-up interview, whether the mother reported at baseline that the father had ever been in prison or jail, and whether the mother reported at baseline that the father has together now?” At follow-up, if the mother indicated she was in a romantic relationship with the baby’s father when the baby was born, she was asked “When [CHILD] was born, were you and [FATHER] living together all or most of the time, some of the time, rarely, or never?” 8 trouble keeping a job or getting along with family and friends because of drug or alcohol problems. The father being unemployed, having a criminal history, or having a substance abuse problem all would reduce his attractiveness as a mate and perhaps make it more likely that the mother will downward revise the baseline relationship. Finally, based on the literature on autobiographical memory discussed earlier, we expect that the path the parents’ relationship has followed since the baseline interview would affect mothers’ retrospective reports. At the follow-up interview, mothers were asked about their current relationship with the baby’s father—whether they were married, romantically involved, separated/divorced, just friends, or not involved, or whether the father was not known. If they were married or romantically involved, they were asked whether they and the father were currently living together all or most of the time, some of the time, rarely, or never. We categorize mothers who were unmarried at follow-up as living together all or most of the time, some of the time, romantic but not living together, or not involved. We also consider whether each parent became involved with a new partner between baseline and follow-up, the quality of the parents’ relationship as characterized by the mother at follow-up (excellent or very good, versus good, fair, or poor), and how much the mother indicated at follow-up that she would trust the father to take care of their child for one week if she had to go away and could not take the child with her (very much, somewhat, or not at all).2 RESULTS We first compare mothers’ follow-up responses to direct questions about their relationship and living arrangement with the father at the time of the birth to their baseline reports. Table 2 presents the raw data for all 4302 mothers who completed follow-up interviews 2 The follow-up questions on relationship quality and trust were asked in 18 cities only. 9 and had non-missing data on the relationship and living arrangement questions. While most respondents fall along the diagonal, a non-trivial number of mothers changed their reports over time, in both directions. In the analyses that follow, we analyze the predictors of changes in reports in each direction in order to uncover potential bias in the retrospective reports. [Table 2 here] The consequences of potential bias in retrospective reports for both estimates of cohabitation and analyses of union formation and dissolution among unmarried parents depend on the analyses that are being conducted as well as on the extent to which inconsistencies represent random recall error. The implication for estimating cohabitation rates depends on the net direction of bias and whether recall is related to the current relationship or other individual characteristics. For example, if couples who break up are more likely to retrospectively revise downward their past relationship and parents who move in with one another after the birth of the child are more likely to retrospectively upgrade their past relationship, then rates of union formation during the year following the birth of the child might be underestimated, even if the rate of baseline cohabitation is not affected. Similarly, if retrospective reports are a function of individual characteristics such as age or race/ethnicity, using them may produce biased estimates of the effects of these characteristics on changes in relationship status. We look at how retrospective reporting varies by sociodemographic characteristics, father attributes, and characteristics of the relationship, as well as the amount of time elapsed between the baseline and follow-up interviews. We limit the sample to respondents who reported at both waves that they were unmarried at baseline. We exclude from most analyses the 290 respondents 10 who reported at follow-up that they were cohabiting with the father some of the time at baseline (corresponding to an effective sample reduction of 277 observations because of missing data on covariates).3 We classify respondents as revising upward (reporting at baseline that they were not cohabiting, but at follow-up that they had been), revising downward (reporting at baseline that they were cohabiting, but at follow-up that they had not), and not revising.4 In Table 3, we present the bivariate associations between parent characteristics and changes in the mothers’ reports of cohabitation, excluding respondents who reported at followup that they cohabited with the father some of the time at baseline. The change in reports is nonnegligible, with 16 percent of unmarried mothers revising their reports of baseline cohabitation status between the first and second waves. Three quarters of those who revised did so in the upward direction (12%, compared to 4% revising downward). This upward drift has implications for the estimation of cohabitation rates and analyses of union formation and dissolution (we return to this later). [Table 3 here] From the bivariate associations, it appears that instability in reports is associated with both sociodemographic attributes of the mother and the quality of the parental relationship. In particular, non-Hispanic white mothers are less likely to revise reports than are non-Hispanic black, Hispanic, and other non-white non-Hispanic mothers; foreign-born mothers are less likely to revise than are US-born mothers; and more educated mothers are less likely to revise than 3 Of these 277 cases, 86 reported at baseline that they were living with the father and 191 reported that they were not. 4 It is important to note that this characterization reflects living arrangements and not necessarily the quality of relationships. 11 mothers with less education. Similarly, young mothers (less than 18) are more likely to revise reports than older mothers, although the relationship is not statistically significant. Consistent with the literature on memory recall, the length of time between interviews increases the likelihood of inconsistent reports. Instability in reports is not associated with father attributes, including whether the father was unemployed, whether he was ever in prison or jail, and whether he had problems related to drugs or alcohol. However, it is strongly associated with relationship status at follow-up. Mothers who were cohabiting all or most of the time with the father at follow-up were more than twice as likely to revise their reports upward, and six times less likely to revise their reports downward, as mothers who were not romantically involved with the father at follow-up. The pattern is clear: Whether consciously or not, mothers tend to reconstruct their past relationships based on current relationship status. This tendency is also evident for other measures of attachment, including whether mothers or fathers have new romantic partners and whether the mother trusts the father with the child, but not for self-assessment of overall relationship quality. The bivariate associations are sustained in multivariate analyses of upward revision and downward revision of baseline relationship status (Tables 4 and 5). Due to collinearity of presence of new partners with parents’ relationship status at follow-up, and missing data on relationship quality and trust from 2 cities (see footnote 2), we excluded these measures from the regression models. In both Tables 4 and 5, Model 1 controls only for sociodemographic variables and number of months between interviews, Model 2 adds father attributes, Model 3 includes sociodemographic measures plus follow-up relationship status, and the last model includes the entire set of covariates. All models exclude respondents who reported at follow-up that they 12 lived with the father some of the time at baseline or had missing values on any of the analysis variables. The effect of race/ethnicity on revising upward is relatively consistent across specifications, as are the effects of nativity and education (Table 4). White mothers and more highly educated mothers are less likely to revise reports upward than are their non-white and less educated counterparts. The effects of nativity and length of time between interviews actually increase slightly when measures of parent relationship are included (Model 3). Father attributes are not significantly associated with revising upward. Finally, follow-up non-cohabitors, particularly those who are no longer involved with the father, are less likely than follow-up cohabitors to revise upward. The magnitudes of these effects are large: Non-Hispanic blacks are over 3 times as likely (e1.14) as whites, foreign-born mothers are 44% as likely as US-born mothers, and mothers who were not cohabiting with the father at follow-up were 30 and 45% (for romantically involved and not involved, respectively) as likely as cohabiting mothers, to revise upward. [Table 4 here] [Table 5 here] The results in Table 5, which show the effects of parent and relationship characteristics on the likelihood of revising downward, are similar to those in Table 4. However, sociodemographic factors generally have weaker effects on revising downward than upward, while the effect of follow-up relationship status is stronger for revising downward. Being nonHispanic black (versus non-Hispanic white) and foreign-born (versus US-born) are insignificant 13 predictors of revising downward, whereas they are strong predictors of revising upward. The effects of education are different for revising upward and downward: Having more than a high school education has a smaller negative effect on revising downward than it does on revising upward, but having less than a high school degree (compared to just a high school degree) is significantly negative only for revising downward. The effect of time elapsed between interviews is similar for both outcomes. The strongest predictor of downward revision is the follow-up relationship status. Relative to mothers who were cohabiting with the father at the time of the follow-up interview, mothers in a romantic relationship with the father at follow-up (but not cohabiting with him) were about 10 times (e2.35) more likely to revise their baseline cohabitation reports downward. Those who were no longer involved with the father were almost 7 times more likely to revise downward. Mothers who reported cohabiting with the father some of the time were 5 times more likely to revise downward.5 Our finding that retrospective reports of cohabitation at birth differ systematically from contemporaneous reports only one to two years earlier raises the question of whether aggregate rates of cohabitation, union formation, and union dissolution based on retrospective reports are biased. As reported earlier, a limitation of the Fragile Families data is that the measure of baseline cohabitation is not exactly the same in the first and second waves. Simply excluding respondents who retrospectively reported living together some of the time at birth could potentially lead to biased results. We therefore provide a range of estimates that treat this group 5 We ran alternative models that, instead of using the sample of mothers who reported at both time points that they were unmarried at baseline, included only mothers who reported cohabitating at baseline at either interview (results not shown). This alternative sample captures revisions of reports to and from marriage in addition to revisions in cohabitation status, but is not representative of unmarried parents. By restricting the sample in this manner, the rates of revision are much higher than those we present, but the predictors of instability in reports of cohabitation are very similar. 14 alternatively as cohabiting, non-cohabiting, excluded from the sample, and apportioned into the two statuses based on cohabitation reports at baseline among these respondents. We consider the last, which assumes that 31% would have reported cohabiting and 69% would have reported not cohabiting had they not been offered the some of the time response category, to be our “best estimate.”6 We present these different retrospective rates in Table 6, along with the cohabitation rate based on the mothers’ contemporaneous baseline reports. Although varying treatment of the some of the time group results in different retrospective estimates, they are uniformly higher than the estimate using contemporaneous reports. Baseline reports yield an estimate of 47.6% unmarried parents cohabiting at the time of the birth, compared to 55% using our best retrospective estimate—that which apportions the some of the time group into cohabiting and non-cohabiting based on the distribution of their baseline contemporaneous reports.7 [Table 6 here] [Table 7 here] Discrepancies between retrospective and contemporaneous reports are even more consequential when estimating rates of union formation (into cohabiting or marital unions), but appear to matter little for estimating rates of union dissolution (Table 7). The analyses of union formation are restricted to mothers who reported not cohabiting at baseline, and those of dissolution are restricted to mothers who reported cohabiting at baseline. The contemporaneous estimates use mothers’ baseline reports of baseline cohabitation status and their follow-up reports 6 Of the respondents who retrospectively reported they were living with the father some of the time, 31% reported at baseline that they were currently living with the father and 69% reported that they were not. 7 These figures, as well as all others we present, are based on unweighted data. 15 of their living arrangement at that time. The retrospective estimates use follow-up reports of both baseline and follow-up cohabitation with the father. The ranges in estimates reflect the different treatments of the some of the time cohabitor group. The estimates of union formation derived from retrospective reports are approximately half those based on contemporaneous reports (10 to 12.6%, compared to 22.5 to 28%). The difference is due to the tendency for mothers who form unions to retrospectively report that they were cohabiting at baseline, which excludes them from the calculations of rates of union formation using retrospective reports. Dissolution rates are somewhat higher using contemporaneous reports, but the differences in rates are much smaller than for union formation because few mothers reporting cohabitation at baseline retrospectively revised their reports downward and calculations are therefore based on very similar samples. Furthermore, those who reported not living together at the time of the follow-up interview were as likely to retrospectively revise baseline cohabitation rates upward as downward. Finally, we estimated multivariate models of union formation and dissolution to assess the sensitivity of the coefficients of race/ethnicity, age, education, and other covariates to whether retrospective or contemporaneous reports of cohabitation are used. We found that the magnitudes of the coefficients of race/ethnicity were particularly sensitive to choice of report. The effect on union dissolution of being non-Hispanic black (relative to being non-Hispanic white) was almost 5 times greater when baseline reports of cohabitation were used than when retrospective reports were used, and the effect of being Hispanic was over three times larger 16 when using retrospective reports compared to baseline reports.8 We also found differences in effect sizes of race/ethnicity when estimating union formation (in the opposite directions). DISCUSSION Despite the relatively short time span between baseline and follow-up interviews in the Fragile Families and Child Wellbeing Study, respondents’ reports of cohabitation sometimes change over time. These changes in reports do not appear to be due entirely to random recall error. Rather, changes in reports over time are associated with individual factors and characteristics of the relationship at follow-up. They may also be a function of changes in interview mode between waves. Due to the shift in mode, it is difficult to tell from these data whether baseline reports of cohabitation are more or less accurate than retrospective reports. Generally, obtaining information about the present is considered more reliable than obtaining information about the past. However, conducting the baseline interviews in-person in the hospital, an institutional setting (versus at home over the telephone at follow-up), may have caused mothers to underreport cohabitation for fear of adverse repercussions. This possible scenario is consistent with anecdotal reports by Fragile Families interviewers that some mothers feared exposing fathers to immigration authorities or the criminal justice system. It is also possible that mothers underreported cohabitation because they thought that this information might affect their eligibility for welfare. Nevertheless, since all study participants were exposed to the same interviewing modes at each wave, the mode effect is unlikely to account for the systematic 8 Using baseline reports, effects (and standard errors) of being non-Hispanic black and Hispanic on union dissolution were .273 (.180) and -.091 (.199), respectively, compared to .059 (.180) and -.354 (.206) when using follow-up reports of baseline cohabitation status (full regression results not shown). 17 associations between changes in reports of cohabitation and both socioeconomic attributes and relationships characteristics. That is, we find evidence of systematic bias in retrospective reports that is not an artifact of the design of the Fragile Families and Child Wellbeing study. Our findings are consistent with theories of Ross (1989), that individuals project current status onto the past, and of Holmberg and Holmes (1994), that current affect toward an individual (the father in this case) affects memory of past relationships. This should raise concerns about retrospective reports of cohabitation in surveys as well as in qualitative studies, as both may be subject to similar physical, social, and psychological forces shaping memory reconstruction. Although we have characterized differences between baseline and follow-up reports as problematic, they may be less consequential for measuring trends in cohabitation and they may be valuable data in their own right. For example, downward revision of baseline cohabitation (from cohabiting to not cohabiting) might be suggestive of the nature of the relationship between the parents at follow-up, which might help to predict future relationship trajectories. Also, the complications associated with reconciling multiple sources of data should not overshadow the benefits of having more information. By including both contemporaneous and retrospective reports of cohabitation as well as indirect measures of cohabitation, surveys can provide researchers with information to more fully characterize family circumstances and dynamics. 18 Table 1. Percent of unmarried mothers cohabiting at baseline, using various measures from baseline reports Living with father Father on household roster Living with father OR father on household roster Living with father AND father on household roster N 48 47 50 45 3632 19 Table 2. Retrospective reports (at 1-year follow-up) by contemporaneous reports of baseline relationship status Baseline Report Not Father Follow-up report: Married Cohabiting cohabiting unknown TOTAL Cohabiting most or all of the time 20 1290 333 3 1646 Cohabiting some of the time 0 88 201 1 290 Not cohabiting 3 105 1062 13 1183 Father unknown 0 0 13 22 Divorced 11 6 30 0 47 TOTAL 1070 1546 1660 26 4302 20 Table 3. Percent revising baseline relationship status, by parent attributes and follow-up relationship status Revised Revised Did not upward downward revise N MOTHER SOCIODEMOGRAPHICS Age < 18 18 3 78 93 18 – 29 12 4 84 2177 30 + 11 4 86 387 Race/ethnicity*** Non-Hispanic white 5 2 93 427 Non-Hispanic black 15 4 81 1452 Hispanic 10 4 87 717 Other 11 8 80 61 Nativity** US-born 13 6 83 2333 Foreign-born 4 3 91 324 Education** Less than high school 13 3 84 1031 High school or GED 14 5 82 909 More than high school 9 4 88 717 # MONTHS BETWEEN INTERVIEWS* <12 8 3 89 274 12 – 17 12 4 85 1755 >17 15 5 81 628 FATHER ATTRIBUTES Employed at follow-up Yes 12 3 85 1789 No 13 5 82 712 Ever jailed Yes 12 4 84 933 No 12 4 84 1582 Drug problems before child’s birth Yes 11 2 86 162 No 12 4 84 2495 MOTHER/FATHER RELATIONSHIP Relationship status at follow-up*** Married 11 1 87 210 Cohabiting most of time 16 1 83 1065 Cohabiting some of time 22 5 73 83 Romantic non-cohabiting 10 9 81 208 Not romantically involved 7 6 87 1090 Relationship quality at follow-up (18 cities) Excellent or very good 15 4 82 1070 Good, fair, or poor 12 4 84 1083 Mother has new partner* Yes 7 4 88 393 No 13 4 84 2251 Father has new partner*** Yes 4 4 92 233 No 13 3 83 2272 Trusts father with child (18 cities)*** Very much 16 3 81 1582 Somewhat 9 5 86 288 Not at all 5 5 90 220 TOTAL 12 4 84 2657 *p < .05, ** p < .01, *** p < .001 21 Table 4. Effects of mother, father, and relationship characteristics on upward revision of baseline relationship status, among mothers who were unmarried at time of birth (logit coefficients) MOTHER ATTRIBUTES Age <18 30+ Race/ethnicity Non-Hispanic black Hispanic Other Nativity Foreign-born Education Less than high school More than high school # MONTHS BETWEEN INTERVIEWS Model 1 Model 2 Model 3 Model 4 0.33 -0.01 0.32 -0.02 0.50 -0.02 0.49 -0.01 1.02*** 0.77** 1.02* 1.02*** 0.77** 1.02* 1.14*** 0.76** 1.14* -0.74** -0.77** -0.84** -0.82** -0.03 -0.44** -0.02 -0.45** 0.03 -0.44** 0.03 -0.43* 0.03 FATHER ATTRIBUTES Employment Unemployed at follow-up Incarceration Ever jailed Drug Drug problems before child’s birth 0.03 2657 0.05** 0.05** -0.00 0.16 -0.14 -0.01 -0.02 0.20 FOLLOW-UP RELATIONSHIP STATUS Married Cohabiting some of the time Romantically involved Not involved N *p < .05, ** p < .01, *** p < .001 1.16*** 0.77** 1.17* 2657 -0.27 0.10 -0.76** -1.16*** -0.26 0.10 -0.80** -1.19*** 2657 2657 22 Table 5. Effects of mother, father, and relationship characteristics on downward revision of baseline relationship status, among mothers who were unmarried at time of birth (logit coefficients) MOTHER ATTRIBUTES Age <18 30+ Race/ethnicity Non-Hispanic black Hispanic Other Nativity Foreign-born Education Less than high school More than high school # MONTHS BETWEEN INTERVIEWS Model 1 Model 2 Model 3 Model 4 -0.11 -0.05 -0.13 -0.03 -0.35 -0.05 -0.34 -0.06 0.70 0.85* 1.66** 0.63 0.81 1.60** 0.38 0.84* 1.41* -0.45 -0.43 -0.28 -0.35 -0.46 -0.24 -0.48 -0.22 -0.53* -0.27 -0.52* -0.29 0.05 0.04 0.06* FATHER ATTRIBUTES Employment Unemployed at follow-up Incarceration Ever jailed Drugs Drug problems before child’s birth 0.06* 0.34 0.08 -0.07 -0.22 -0.45 -0.67 FOLLOW-UP RELATIONSHIP STATUS Married Cohabiting some of the time Romantically involved Not involved N *p < .05, ** p < .01, *** p < .001 0.44 0.87* 1.45* 2657 2657 0.43 1.66** 2.33*** 1.89*** 0.42 1.65** 2.35*** 1.94*** 2657 2657 23 Table 6. Estimates of cohabitation at birth using mothers’ baseline and follow up reports Percent cohabit Baseline reports Follow up reports: Including “some of the time” as cohabitors Including “some of the time” as non cohabitors Excluding “some of the time” Apportioning “some of the time” 31/69* * Represents best estimate from follow-up reports 47.6 61.4 52.1 57.5 55.0 24 Table 7. Rates of union formation and dissolution, using mothers’ contemporaneous and retrospective reports Union formation (among baseline non cohabitors) Union dissolution (among baseline cohabitors) Contemporaneous 22.5 to 28.9 25.4 to 29.7 Retrospective 10.0 to 12.6 23.1 to 26.4 25 REFERENCES Baughman, R., Dickert-Conlin, S. and Houser, S. 2002. How well can we track cohabitation using the SIPP? A consideration of direct and inferred measures. Demography, 39(3): 455-65. Binstock, G. and Thornton, A. 2003. Separations, reconciliations, and living apart in cohabiting and marital unions. Journal of Marriage and Family. 65(2):432-443. Bower, G.H. 1987. Commentary on mood and memory. Behavior Research and Therapy, 25, 443-456. Bumpass, L. and Lu, H. 2000. Trends in cohabitation and implications for children’s family contexts in the U.S. Population Studies, 54: 29-41. Bumpass, L. and Raley, K. 2003. Measuring divorce and separation: Issues, and comparability of estimates across data sources. Paper presented at the conference on Measurement Issues in Family Demography in Bethesda, MD, November 13-14. Bumpass, L. and Sweet, J. 1989a. National estimates of cohabitation: cohort levels and union stability. Demography, 26: 615-625. Bumpass, L. and Sweet, J. 1989b. Children’s experience in single-parent families: Implications of cohabitation and marital transitions. Family Planning Perspectives, 21(6): 256-61. Casper, L.M. Cohen, P.N. 2000. How Does POSSLQ measure up? Historical estimates of cohabitation. Demography, 37: 237-245. Edin, K. and Lein, L. 1997. Making Ends Meet: How Single Mothers Survive Welfare and LowWage Work. New York: Russell Sage Foundation. Groves, R. 1989. Survey Errors and Survey Costs. New York: Wiley. Fields, Jason and Martin T. O'Connell. 2003. Direct and indirect measures of cohabitation in census bureau data: Who's just sharing rents? Paper presented at the 2003 meeting of the Population Association of America. Hasher, L., and Griffin, M. 1978. Reconstructive and reproductive processes in memory. Journal of Experimental Psychology: Human Learning and Memory, 4: 318-330. Holmberg, D., and Holmes, J. G. 1994. Reconstruction of relationship memories: A mental models approach. In N. Schwarz, & S. Sudman (Eds.), Autobiographical memory and the validity of retrospective reports. New York: Springer-Verlag. 26 Isen, A.M. 1987. “Positive affect, cognitive processes, and social behavior,” In L. Berkowitz (Ed.), Advances in Experimental Social Psychology. 20:203-253). San Diego,CA: Academic Press. Lillard, L. and Waite, L. 1988. Panel versus retrospective data on marital histories: Lessons from the PSID. Paper presented at the Social Science Research Council Conference in Annapolis, Maryland, "Individuals and Families in Transition: Understanding Change Through Longitudinal Data." Compiled by Hazel V. Beaton, Debra A. Ganni, and Delma T. Frankel. US Department of Commerce, Bureau of the Census. Lin, I., Schaeffer, N.C., Seltzer, J. and Tuschen, K. (forthcoming) Divorced parents’ qualitative and quantitative reports of children’s living arrangements. Journal of Marriage and Family. Loftus, E.L. and Marburger, W. 1983. Since the eruption of Mount St. Helene’s, has anyone beaten you up? Improving the accuracy of retrospective reports with landmark events. Memory and Cognition, 11:114-20. Manning, W.D. 1993. Marriage and cohabitation following premarital conception. Journal of Marriage and the Family, 55: 839-851. Manning, W.D. and Smock, P.J. 2003. Measuring and modeling cohabitation: New perspectives from qualitative data. CFDR Working Paper 2003-10. McLanahan, S., Garfinkel, I., Reichman, N., Teitler, J. 2001. Unwed parents or fragile families? Implications for welfare and child support policy. In Lawrence Wu and Barbara Wolfe (eds.), Out of Wedlock: Causes and Consequences of Non-Marital Fertility. Russell Sage Foundation. Nock, S. L. 1995. A comparison of marriages and cohabiting relationships. Journal of Family Issues, 16: 53-76 Peters, E. 1988. Retrospective Versus Panel Data in Analyzing Lifecycle Events. Journal of Human Resources, 23(4): 488-513. Reichman N., Teitler J., Garfinkel I., McLanahan, S. 2001. Fragile Families: Sample and design. Children and Youth Services Review, 23(4): 303-326. Rips, L., Conrad F. and Fricker, S. 2003. Seam effect in panel surveys. Public Opinion Quarterly, 67(4): 522-554. Ross, C.E. 1995. Re-conceptualizing marital status as a continuum of social attachment. Journal of Marriage and the Family, 57: 129. Ross, M. 1989. Relation of implicit theories to the construction of personal histories. Psychological Review, 96: 341-357. 27 Schaeffer, N.C. 1995. Errors of experience: Response errors in reports about child support and their implications for questionnaire design. Schwartz, N. and Sudman, S. (eds.) Autobiographical Memory and the Validity of Retrospective Reports. Springer-Verlag. New York. Seltzer, J.A. 2000. Families formed outside of marriage. Journal of Marriage and the Family, 62: 1247–1268. Shum, M. and Ripps, L. 1999. The respondent’s confession: Autobiographical memory in the context of surveys. In Cognition and Survey Research, ed. Serkin, M., Herrmann, D., Schechter, S., Schwarz, N., Tanur, J., and Tourangeau, R., pp. 95-109. New York: Wiley. Smock, P.J. 2000. Cohabitation in the United States: An appraisal of research themes, findings, and implications. Annual Review of Sociology, 26: 1-20. Teitler, J. and Reichman, N. 2001. Cohabitation: An elusive concept. Social Indicators Survey Center Working Paper #01-4 and Center for Research on Child Wellbeing Working Paper #2001-07. 28