2003 Joint Statistical Meetings - Section on Government Statistics Improving Quality in the Collection of Earnings: The Survey of Income and Program Participation 2004 Panel Nancy Bates and Aniekan Okon U.S. Census Bureau, Demographic Surveys Division, Washington, DC 20233-8400 randomly assigning each to a treatment group or a control group. Each household in the treatment group received a modified SIPP instrument containing experimental questions. Each household in the control group received the current, standard SIPP instrument. Interviewers who conducted the interviews were all experienced SIPP interviewers who received special training on the new, experimental questions and procedures. Two more field tests with similar designs were conducted in 2001 and 2002, each with two separate waves. (For a more comprehensive background on the SIPP Methods Panel, see Doyle, Martin and Moore, 2000.) KEY WORDS: Labor Force, Earnings, Nonresponse, Reporting options 1. Abstract1 In an attempt to reduce respondent burden and item nonresponse to income questions, the Survey of Income and Program Participation (SIPP) will implement new procedures to collect earnings from work activities in the 2004 panel. In previous SIPP instruments, the primary method of reporting employer earnings has been in monthly amounts; when gathering business earnings, monthly is the only option allowed. The 2004 panel will allow respondents a choice in how they report earnings from businesses and employers. Prior to a battery of questions about earned income amounts, interviewers probe respondents for their preferred time period of reporting earnings—either annually, hourly, biweekly, bimonthly, monthly, or quarterly. The new SIPP instrument also adds several probes in the earnings section aimed at reducing item nonresponse and introduces dependent-style interviewing techniques during the Wave 2+ interviews. In this paper, we discuss how the new methods impacted item nonresponse and data quality in a series of field tests. In this paper, we describe new techniques designed to encourage better income reporting in the 2004 SIPP. In particular, from the Wave 1 questionnaire, we detail a new method of flexible reporting periods, expanded use of computerized verification checks, automated monthly transformations, and new nonresponse follow-up probes. From the Wave 2 interview, we describe the use of dependent interviewing as a means of reducing nonresponse to questions about earnings from wage and salary jobs and businesses. We then present a comparison analysis of item nonresponse metrics and mean and median reported earnings between the new 2004 SIPP Panel and the current SIPP instrument. We end with a brief discussion of how changes in the 2004 SIPP will impact data output and then summarize conclusions from the Methods Panel field tests. 2. Introduction The 2004 Survey of Income and Program Participation (SIPP) Panel has gone through significant instrument redesigns since the inception of the SIPP Methods Panel. The SIPP Methods Panel was established as a means of designing and testing changes to the 2004 SIPP during a series of production-like field tests. Some overriding goals of the 2004 SIPP are to increase efficiency and accuracy, reduce redundancy, develop “friendlier” wording, simplify response tasks, and, in general, reduce respondent burden (Moore et al. 2002). 3. Labor Force Income Questions Like all sections of the core SIPP instrument, the 2004 labor force section has undergone a program of changes and testing. We limit the improvements discussed in this paper to the Labor Force 2 section of the SIPP. This section of the core instrument collects detailed information about earnings from wage and salary jobs and businesses, profits from businesses, earnings from tips, commissions, overtime and bonuses, “other” business income, contingent work income, and moonlight earnings.2 Our experience with the new techniques is divided into results from the Wave 1 and Wave 2+ instruments. In addition to the broad methodological changes discussed below, we note that the 2004 labor force section also contains several new questions and imposes different universes for old questions in some instances. These are not discussed in detail here but are referenced a few places in the paper and briefly discussed in Section 5. The Methods Panel project carried out its initial field test in August and September of 2000. The 2000 field test drew a representative sample of households in six of the Census Bureau's twelve regional offices—Philadelphia, Kansas City, Seattle, Charlotte, Atlanta, and Dallas— 1 This paper reports the results of research and analysis undertaken by Census Bureau staff. It has undergone a Census Bureau review more limited in scope than that given to official Census Bureau publications. This report is released to inform interested parties of ongoing research and to encourage discussion of work in progress. 3.1 Wave 1 - Reporting Options 2 The core SIPP routinely asks all detailed earnings questions about the 1st and 2nd job or business only. 462 2003 Joint Statistical Meetings - Section on Government Statistics “Earlier you stated that you usually work 40 hours per week. At $8 per hour, that works out to about $320 every week. Does that sound about right for your gross income?″ When reporting job earnings in the traditional SIPP, respondents are asked: Each time you were paid by [Employer Name] in [Month X], how much did you receive BEFORE deductions? The instrument then uses weekly gross amounts to calculate the desired final outcome -- gross monthly amounts. This method takes advantage of computer-assisted automation by allowing flexibility in reporting options and putting more of the income calculation burden on the computer, not the respondent. Our intent was to increase the validity and accuracy of the data being collected and reduce respondent burden at the same time. Interviewers are instructed to enter one or more gross amounts for the reference month. This allows separate paycheck amounts or a single aggregated monthly amount to be entered. There is also an option to hit “A” and enter an annual amount if the respondent reports being paid annually in response to the question. Likewise, the instrument also has an option to hit “C” and calculate a monthly amount by entering an hourly wage and a number of hours worked. If hourly is selected, a monthly amount is calculated and read back to the respondent to verify. However, neither the hourly nor the annual option is explicitly offered to the respondent as part of the question. For salaried income from a business, gross monthly amount is the only reporting option allowed. 3.2 Wave 1 – Item Nonresponse When revising the SIPP 2004 labor force earnings section, another of the goals was to reduce item nonresponse. Item nonresponse occurs when a respondent refuses or is unable to answer a particular question. This may occur in cases where the reporting task is too difficult, the information is not available, or the question is deemed too sensitive. This is a fairly common occurrence for income questions in the case of a proxy reporter or in cases where answers are withheld because they are viewed as too private or inappropriate to be shared with a stranger. Like other income surveys, the SIPP has experienced substantial levels of item nonresponse to income questions (Dixon et al. 2003, Atrostic and Kalenkolski 2002, Roemer 2000). The method in the 2004 instrument took a very different tact—it is flexible and allows respondent-defined periods for reporting job earnings and salaried business income. At the beginning of the earnings section, respondents are informed that the goal is to collect gross monthly earning amounts. However, the instrument allows respondents to select their preferred periodicity of reporting: monthly, biweekly/bimonthly, annually, hourly, or quarterly (in the case of businesses). If none of these are selected, respondents may report “some other way”, in which case the instrument prompts for monthly amounts. The exact question wording is as follows: In an effort to reduce item nonresponse, the 2004 SIPP instrument instituted a two-tiered approach for nonresponse follow-up probes during the Wave 1 interview. At the first level, if the respondent answers “don’t know” or refuses an income amount question, the respondent is asked to provide an approximate monthly, biweekly or annual amount depending upon the type of reporting method chosen. The goal of this part of the survey is to find out about your MONTHLY GROSS income from this job, BEFORE taxes and other deductions. We can do this in several ways. We can go straight to monthly totals, or we can try to work with [FILL - weekly/biweekly/bimonthly] paychecks, hourly pay rates, or annual income amounts if that would be easier. What would be easiest for you? The second level occurs if the respondent still answers “don’t know” at the approximate income amount screen. The respondent is then asked to try and report a gross monthly amount (the default method). The specific wording is: The rationale behind the new flexibility is to make retrieval and reporting more natural and in line with how respondents typically think about earnings. Although the goal of the SIPP is to collect gross monthly income, respondents commonly must use “calculating” techniques to arrive at these amounts; e.g., recalling number of hours worked per week and multiplying times pay rates and number of weeks worked per month (U.S. Census Bureau 2000). “Let’s try another way. How much did you receive BEFORE deductions from [Employer Name] for [Month X]?” If the respondent refuses at the default monthly screen, the instrument skips over the labor force earnings questions. It was our intent to reduce item nonresponse by encouraging approximate reporting first and then trying one last time to collect information by the standard monthly method. In cases where an amount other than monthly is selected, the SIPP 2004 instrument internally calculates a gross monthly amount based on pay dates, pay periods, hours worked, paycheck totals, annual salaries, and other pertinent information necessary to make a calibrated monthly transformation. It then verifies the amount to make sure the calculation matches up with what the respondent has in mind. For example, after a respondent reports income as hourly wages, the instrument verifies by asking: 3.3 Wave 2- Dependent Interviewing In Waves 2+ of the SIPP, most sections of the instrument repeat questions from the first interview in order to measure change over time. This concept is at the heart of longitudinal surveys and critical for measuring the economic and social condition of important subpopulations over time. As part of the 2004 SIPP redesign, the decision was made to maximize 463 2003 Joint Statistical Meetings - Section on Government Statistics the use of dependent interviewing in the labor force and other sections of the instrument. (In this paper, “dependent interviewing” refers to the use of information from prior interviews, preloaded into the current computer-assisted data record. It does not refer to the use of information from administrative records or other sources.) respondent is probed for the correct amount. The decision to use dependent interviewing in this manner was based on the desire to maximize the reporting of change between waves—we feared that presenting previously reported amounts up front might encourage respondents to acquiesce and accept previous wave data to shorten the interview. The use of dependent interviewing is generally viewed as a more efficient approach to collection of information, since interviewers do not need to repeat questions from previous waves (Mathiowetz and McGonagle 2000). A few published studies indicate that the use of this technique leads to an improvement in data quality by reducing reporting bias and increasing item response (Hill 1994, Dibbs et al. 1995). The survey methods literature categorizes dependent interviewing into two techniques: “proactive” and “reactive.” (See Brown et al. 1998.) In proactive dependent interviewing, the respondent is first reminded of information acquired from a previous interview and then asked about their current situation, often by verifying whether previous information is still accurate. With reactive dependent interviewing, a question from the previous interview is re-asked during the current interview and information from the previous interview is used to detect discrepancies between current and previous answers. An example of the dependent interviewing path in the SIPP 2004 labor force earning section is illustrated below: >EASYWAY< “The next questions are about your gross income from [employer name], BEFORE taxes and other deductions. Last time, we used annual amount to report amounts. Is that still a good way to go?” If Yes, go to AMOUNT. >AMOUNT< “What is your current salary?” If Don’t Know or Refused and an annual amount was reported in the last wave, go to PWAMT1 If Don’t Know or Refused and no annual amount is available from last wave, go to APXAMT1 The 2001 Wave 2+ SIPP currently uses proactive dependent interviewing in the section where information about labor force participation is collected; however, no form of dependent interviewing is used in the section that collects income amounts from jobs or businesses. This changes in the 2004 SIPP, which uses a combination of proactive and reactive interviewing. >PWAMT1< “Things may have changed, but I have recorded from the last time that your annual income from this job was about [$ previous wave amount]. Does that still sound right?” If Yes, record previous wave amount into current wave. If No, go to PWAMT1FIX If Don’t Know, go to TRYMNTH If Refused, record refused into current wave amount. In cases where the information is available from the previous interview, respondents are reminded of the reporting method they previously selected and asked if this is still an acceptable way to report income (proactive method).3 However, when reporting actual amounts, we elected to use dependent interviewing only when respondents failed to provide an answer to questions about income amounts from jobs and businesses (don’t know or refused). In such cases (and where information from the previous interview is stored and available), the instrument pulls up a screen that reminds the respondent of amounts reported previously and asks the respondent to verify if this is still accurate (reactive interviewing). If it is, the instrument stores the previous wave amount into the current wave amount field and the interview proceeds. If the amount is not accurate, the >PWAMT1FIX< “What is the correct amount?” If Don’t Know, go to TRYMNTH If Refused, record refused into current wave amount.. >APXAMT1< “Can you give me an approximate amount?” If Don’t Know, go to TRYMNTH If Refused, record refused into current wave amount. 3 Preloading information from a previous wave and feeding it back to the current wave is guided by the Census Bureau’s Respondent Identification Policy (RIP). This policy guides the sharing of personal information within a household (Gates 1998). Without approval from the respondent, information from the last wave cannot be brought back and used in cases where the respondent in Wave 2 is different from respondent in Wave 1. For the purpose of this paper, we assume that the respondent has granted that permission. >TRYMNTH< “Let’s try monthly amounts. How much did you receive BEFORE deductions from [Employer Name]”? If Don’t Know or Refused, record DK / R into current wave amount. In the 2004 Wave 2+ instruments, all of the labor force 464 2003 Joint Statistical Meetings - Section on Government Statistics individual nonresponse ratios across all adults by the different earnings sections are summarized in Table 2. Again, the table is arranged by replicate and by wave within replicate. The Total Persons Earnings row represents the aggregate nonresponse ratios for all of the labor force earning items combined. earnings screens also contain a feature which essentially allows respondents to self-select a dependent amount question. The “L” response option can be entered if the respondent says something like “whatever I said during the last interview.” If the dependent option turns out not to be available—either because there was no amount from the prior wave or there is a respondent identification problem— the instrument issues a subtle apology: “I am sorry. That information doesn’t seem to be in my computer.” It then proceeds to a Wave 1-type follow-up probe question, asking for approximate amount. For all waves in each replicate, the new instrument yielded significantly lower levels of missing data for items dealing with person-level labor force earnings.4 For example, in Replicate 2 Wave 1, the new instrument had an average ratio of 0.11 missing items compared to 0.17 in the standard SIPP. The improvements are especially noticeable in Waves 2 (0.07 compared to 0.24 and 0.11 compared to 0.20). We surmise that the new dependent interviewing methods were responsible for the decreased nonresponse during the Wave 2 interviews. 4. Results 4.1 Reporting Options Table 1 illustrates the distribution of how respondents reported income amounts from wage and salary jobs and businesses. Results from the 2001 and 2002 replicates are presented and separated by wave within each replicate. Across replicates and waves, the results are very similar: the majority of respondents to the new SIPP instrument (noted as TEST in the tables) chose to report wage and salary job earnings in a manner other than monthly. Over one-third typically chose to report by annual amounts, around 30 percent elected to report earnings via weekly/biweekly/bimonthly paychecks, and around 10 percent used the hourly pay rate option. Less than 20 percent selected to report using monthly aggregate amounts. In all four field tests, the rate of nonresponse to the wage and salary items was smaller in the new instrument than in the control (in all but one, these differences were statistically significant). For three of the field tests, we see lower rates of nonresponse to the business earnings questions—two of which are significantly different: Replicate 2 Wave 2 and Replicate 3 Wave 1. This may be the result of restricting the business earnings questions in the 2004 instrument to those respondents who indicated a salary draw and/or work for an incorporated business. The intent was to tailor these questions to a more appropriate universe and reduce potential confusion and extra reporting burden on business owners whose primary earnings come from profits. Nonresponse ratios for moonlight earnings and net profits/losses were not significantly different between the new and old instruments. As a whole, we view the reductions in nonresponse as another major achievement for the new instrument, particularly since missing data for this section of the survey is historically substantial. These distributions look very different from the 2001 SIPP (noted as CONTROL in the tables). As mentioned previously, while the old SIPP instrument allows wage/salary income to be reported in hourly pay rates or annual amounts, it does not exactly encourage it; that is, the hourly and annual options are not read aloud to the respondent. This is evidenced by the fact that the majority of 2001 SIPP instrument respondents reported monthly amounts (around 80 percent across replicates and waves). In the case of reporting salaried business earnings, monthly is the only option allowed—in the new instrument where other options exist, respondents reported most often by annual amounts followed by monthly totals. We believe these findings are strong evidence that the new ‘periodicity’ reporting method is a big improvement and major accomplishment toward the goal of adopting a more natural, efficient, and respondent-friendly questionnaire. It takes advantage of computer-assisted automation by allowing flexibility in reporting options. 4.3 Amounts The final tables in our analysis examine monthly mean earnings across the old and new SIPP instruments. We present two tables: Table 3 reflects total person labor force earnings; Table 4 represents only business earnings. Table 3 combines all income reported from wage and salary jobs (1st and 2nd jobs), contingent jobs, salary from businesses (1st and 2nd businesses), business profits, other business income, and moonlight income. For the 2004 instrument, this also includes income from new questions that specifically ask about tips, commissions, bonuses or overtime and income from 3rd+ jobs or 3rd+ businesses. (In the old SIPP, these amounts were presumably collected by probes following each monthly amount question.) The combined business earnings reflect only income reported from incorporated or 4.2 Item Nonresponse For each labor force section, we computed a nonresponse ratio for each adult, in which the numerator was the number of “don’t know” and “refused” responses to an income amount question and the denominator was the total number of such questions administered during that person’s interview. We constructed this indicator to examine differences in the degree of item-level “missingness” between the old and new SIPP instruments. The averages of 4 T-tests were performed to detect significant differences between the test and control panels. A sample design effect is factored when performing the tests. See Rottach (2003) for documentation. Tests are based on weighted data. 465 2003 Joint Statistical Meetings - Section on Government Statistics The goal of the SIPP Methods Panel was to produce a new instrument for the 2004 SIPP that was streamlined, efficient, respondent-friendly, and capable of collecting higher quality income data. To achieve this goal, we incorporated new methods—including flexible respondent-identified reporting options, new follow-up probes to encourage response, and the liberal use of both proactive and reactive dependent interviewing. We took advantage of computer-assisted automation by placing more calculation burdens on the computer and fewer on the respondent. salaried businesses and income reported as net profits from a business. Table 3 illustrates the mean combined person earnings by replicate, wave, panel and month. Values in parentheses reflect the maximum values for each month. These are presented to illustrate outliers that may have impacted mean values. Table 3 suggests that, despite the very different data collection methods in the SIPP 2004, the total person earnings are very comparable to the monthly amounts collected in the old SIPP instrument. In fact, no significant differences were found in mean amounts for any of the four reference months in any of the field tests. This is reassuring and suggests that, despite all the revisions in the labor force section, the new SIPP instrument should produce estimates similar to those collected during previous SIPP panels. To evaluate whether our changes were successful, we examined the distribution of reporting options for wage and salary and business income, levels of item-nonresponse for earning questions, and the monthly mean income amounts for the various components of labor force income. Our results were consistent across the four field tests. Respondents reacted favorably to the non-monthly reporting options available for the first time in the 2004 instrument (namely, paycheck amounts and annual salary options). This new method, coupled with the use of dependent interviewing in Wave 2, also paid dividends in terms of item response: Compared to the traditional SIPP instrument, the 2004 version consistently had a lower rate of missing data for person earnings questions. We view this as a major achievement, since labor force earnings constitute the lion’s share of household income yet often suffer from non-trivial levels of nonresponse. Turning to business income amounts, with the exception of Replicate 2 Wave 1, there were no significant differences in mean monthly amounts between the old and new instruments. The difference in Month 2 from this test is likely the result of a large outlier case ($173,333) in the new instrument sample. In Replicate 3, there are several months where business means are noticeably larger for the old SIPP, but the differences are not statistically significant. These months also had large outlier values which obviously influence the mean values. In these cases, however, the medians are actually quite comparable between the old and new instrument (not shown). At the same time, however, the dollar amounts collected using the new method were (for the most part) no different from the monthly amounts reported in the traditional SIPP instrument. The total monthly mean and median person earnings were equivalent across all four field tests, suggesting that income estimates will not be effected by the new design changes. The new SIPP panel will go into full production in January 2004. Based on our extensive research and testing during the Methods Panel field tests, we feel confident our goal of improving income data quality for the SIPP will be realized. 5. Impact of Changes to the SIPP Data Our field tests suggest that revisions to the SIPP instrument will yield better quality data in the form of lower item nonresponse but, at the same time, should not significantly alter employment income estimates. However, because of format changes, universe changes, and new questions and question logic skips, the 2004 instrument output will be different in some respects from the 2001 panel data. For example: In several cases, the universe for a question is smaller for SIPP 2004 because answers can be logically inferred from a previous question. An example of this is a respondent who works in a family business without pay or indicates he does only unpaid work. In these cases, if the respondent indicates he had weeks not worked during the reference period, the instrument will skip the question about whether he was paid for those weeks not worked and instead pre-fill “no” to the question. The universe for the output will not change because the instrument will automatically pre-fill the answer for the newly skipped item. Thus, many of the differences will be transparent to data users because the new data will be mapped back to the old before data files are made available to the public. Detailed information about new items or changed items that impact question content, universes or values is documented in a CD-ROM. (See Doyle, Eargle and Villa, 2003.) 7. References Atrostic, B.K., and Kalenkoski, C. (2002), “Item Response Rates: One Indicator of How Well We Measure Income.” Proceedings of the American Statistical Association, Survey Research Methods Section [CD-ROM], Alexandria, VA: American Statistical Association, pp. 94-99. Bates, N. (2003), “Results of SIPP Methods Panel Labor Force Sections Replicate 3 Wave 1.” Internal U.S. Census Bureau Memorandum from Nancy Bates to SIPP Executive Committee. Brown, A., Hale, A., and Michaud, S. (1998), “Use of Computer Assisted Interviewing in Longitudinal Surveys.” Chapter 10 (pp. 185-200) in Couper, M., R. Baker, J. Bethlehem, C. Clark, J. Martin, W. Nicholls, and J. O’Reilly (eds.), Computer Assisted Survey Information Collection, 6. Conclusions 466 2003 Joint Statistical Meetings - Section on Government Statistics New York: John Wiley & Sons. Testing Results (Replicate 1, Round 1) - Section E: Earnings.” February, 29. Unpublished report. Dibbs, R., Hale, A., Loverock, R., and Michaud, S. (1995), “Some Effects of Computer Assisted Interviewing on the Data Quality of the Survey of Labour and Income Dynamics.” Proceedings of the International Conference on Survey Measurement and Process Quality, Alexandria, VA: American Statistical Association, pp.174-177. Dixon, J., Doyle, P., Eargle, J., Griffin, D., McGovern, P. (2003), “Quality at the Item Level: Terms, Methods and Guidelines for Cross-Survey Comparisons - A Summary.” Paper presented at the Federal Committee on Statistical Methodology Research Conference, November 18, Arlington, VA. Doyle, P., Eargle, J., and Villa, C. (2003), “Survey of Income and Program Participation- A New Instrument for 2004.” Paper prepared for the annual JSM Conference of the American Statistical Association. San Francisco, CA, August 3-7 2003. Doyle, P., Martin, E., and Moore, J. (2000), “The Survey of Income and Program Participation (SIPP) Methods Panel Improving Income Measurement.” SIPP Working Paper No. 234, November 13. U.S. Bureau of the Census. Gates, G. (1998), “Respondent Identification Policy.” Internal U.S. Census Bureau Memorandum from Gerald Gates to Chet Bowie, August 6. Gates, G. (1998). “Respondent Identification Policy.” U.S. Bureau of Census internal memorandum. August 6, 1998. Hill, D.(1994), “The Relative Empirical Validity of Dependent and Independent Data Collection in a Panel Survey.” Journal of Official Statistics, Vol. 10, 359-380. Mathiowetz, N., McGongale, K. (2000), “An Assessment of the Current State of Dependent Interviewing in Household Surveys.” Journal of Official Statistics, Vol. 16, no. 4, pp. 401-418 Moore, J., Pascale, J., Doyle, P., Chan, A., and Griffiths, J. (2002), “The SIPP Methods Panel Project: Using Field Experiments to Improve Instrument Design.” Monograph Paper for the International Conference on Questionnaire Development, Evaluation and Testing Methods (QDET), Charleston, SC, November 14-17 2002. Roemer, M. I. (2000), “Assessing the Quality of the March Current Population Survey and The Survey of Income and Program Participation Income Estimates, 1990-1996.” Housing and Household Economics Statistics Division, June, U.S. Census Bureau, Washington D.C. Rottach, R. (2003), “Design Effects for the SIPP Methods Panel (MPSIPP).” Internal U.S. Census Bureau Memorandum from R.A. Rottach to the Record, June 25. U.S. Census Bureau (2000), “SIPP Methods Panel Cognitive 467 2003 Joint Statistical Meetings - Section on Government Statistics ble 1. Preferred Methods of Income Reporting by Respondents in SIPP Methods Panel by Panel by Replicate Wages/Salary Business Replicate 2 2001 Wave 1 Control Replicate 3 2002 Wave 2 Test Control Wave1 Test Control 77% Test Replicate 2 2001 Wave2 Wave 1 Replicate 3 2002 Wave 2 Test Control Test Wave 1 Control Test Wave 2 Control Test Control Control Test 17% 79% 15% 100% 24% 100% 38% 100% 43% 100% 37% 31% N/A 30% N/A 12% N/A 8% N/A 10% N/A 10% nthly Totals 83% 18% 86% 20% eekly/bimonthly paychecks N/A 30% N/A 30% rly Pay Rates 5% 10% 2% 9% 5% 13% 3% 11% N/A 0 N/A 1% N/A 0 N/A 0 rterly amount N/A N/A N/A N/A N/A N/A N/A N/A N/A 0 N/A 11% N/A 1% N/A 3% ual amount 12% 36% 12% 36% 18% 31% 19% 37% N/A 49% N/A 31% N/A 41% N/A 43% e of the Above N/A N/A N/A N/A N/A 3% N/A 2% N/A N/A N/A N/A N/A N/A N/A N/A N/A 6% N/A 4% N/A 6% N/A 6% N/A 14% N/A 12% N/A 4% N/A 7% 1219 964 1148 879 1516 1367 1429 1152 152 65 126 92 92 69 154 73 N/A Table 2. Labor Force II Aggregate Non-Response Metrics in SIPP Methods Panel by Panel by Replicate Replicate 2 2001 Wave 1 CONTROL Replicate 3 2002 Wave 2 TEST CONTROL Wave 1 TEST CONTROL Wave 2 TEST CONTROL TEST Total Persons Earnings .17 .11** .24 .07** .19 .16* .20 .11** Wage/Salary .16 .06** .23 .05** .16 .14 .19 .09** Business Earnings .19 .23 .25 .03** .34 .19* .27 .18 Moonlight/Odd jobs/Self Employment .07 .20 .24 .21 .04 .03 .10 .14 Net Profit/Loss .39 .38 .34 .33 .35 .36 .32 .39 *Significant difference at .10-level. **Significant difference at .05-level. T-tests used to detect differences between ratios. 468 2003 Joint Statistical Meetings - Section on Government Statistics Table 3. Mean Combined Person Earnings in SIPP Methods Panel by Month by Panel by Replicate Replicate 2 2001 Control Replicate 3 2002 Wave 1 Wave 2 Wave 1 Wave 2 Mean Mean Mean Mean Test Control Test Control Test Control Test Monthly total 1 $3485 (40,000) $3201 (54,167) $3159 (38,460) $3158 (40,000) $3360 (90,000) $3166 (35,833) $3162 (53,000) $3226 (41,658) Monthly total 2 $3371 (49,000) $3479 (173,333) $3219 (38,460) $3165 (40,000) $3211 (70,000) $3160 (33,333) $3148 (53,000) $3244 (41,658) Monthly total 3 $3402 (141,400) $3214 (40,000) $3271 (60,000) $3219 (40,000) $3256 (80,000) $3124 (33,333) $3247 (130,000) $3232 (41,658) Monthly total 4 $3538 (70,000) $3209 (54,167) $3267 (68,000) $3175 (40,000) $3259 (90,000) $3197 (91,657) $3351 (70,000) $3266 (41,658) Table 4. Mean Combined Person Business Earnings in SIPP Methods Panel by Month by Panel by Replicate (Monthly maximum values denoted in parentheses) Replicate 2 2001 Replicate 3 2002 Wave 1 Wave 2 Wave 1 Wave 2 Mean Mean Mean Mean Control Test Control Test Control Test Control Test Monthly total 1 $4217 (25,000) $6156* (23,333) $3950 (30,000) $4324 (15,000) $8191 (100,000) $4746 (20,000) $4963 (30,000) $4456 (20,833) Monthly total 2 $4359 (27,500) $9480* (173,333) $3450 (30,000) $4213 (15,000) $5771 (70,000) $4680 (20,833) $4778 (30,000) $4604 (20,833) Monthly total 3 $3773 (16,500) $5814* (23,333) $4453 (60,000) $4200 (15,000) $6016 (80,000) $4619 (25,000) $8884 (130,000) $4503 (20,833) Monthly total 4 $4719 (62,000) $5888 (25,000) $3945 (30,000) $4098 (15,000) $8386 (100,00) $4697 (20,833) $5198 (30,000) $4671 (20,833) *Significant difference at .10-level. **Significant difference at .05-level. T-tests used to detect differences in mean values. 469