3

Role of the Contingent Worker Supplement in Fulfilling Measurement Needs Related to Alternative Work Arrangements

Although a number of data sources contribute to the measurement and monitoring of labor market trends involving alternative work arrangements (AWAs), the Contingent Worker Supplement (CWS) to the Current Population Survey (CPS) is the key instrument within the portfolio of federal economic statistical programs. This chapter presents recommendations intended to help guide future iterations of this important survey. It includes higher-level, conceptual recommendations as well as more detailed recommendations for improving the survey instrument.

This chapter’s subsections address the universe of workers and jobs covered in the survey (3.2); the categorical distinctions among work arrangements it is most important for the survey to make (3.3); and the characteristics of AWA jobs it is important to identify (3.4). In the existing survey, the CWS respondents are asked only about their main jobs, as identified in the basic CPS questionnaire. As detailed in this chapter, the panel recommends broadening that scope to include other activities a person may have engaged in for pay. Whereas the current survey measures several different AWAs, the most important work-arrangement distinction to make is that between independent contractors and other self-employment arrangements, on the one hand, and traditional employee arrangements on the other. In addition, it is important to capture key characteristics of jobs, most notably the predictability of workers’ schedules as well as the contingency or temporary nature of jobs. Focusing on the characteristics of jobs—and the preferences of workers in relation to those characteristics—would yield more valuable information than attempting to assign jobs to labeled categories. Survey design issues are discussed throughout the

chapter. We begin with an overview of the current structure of the CWS, which is helpful to have in mind when considering what to keep and what to change in the survey.

3.1. OVERVIEW OF THE CONTINGENT WORKER SUPPLEMENT

This section briefly describes the purpose and current structure of the CWS.¹ The CWS collects information from individuals identified as employed in the CPS concerning their main job and accepts both self-reports and proxy responses. Administered by the Census Bureau using a probability sample of about 60,000 households, the CPS asks respondents about activities that took place during the week including the 12th of the month, which for most respondents is the week prior to the survey interview. The interviews are conducted by well-qualified field staff either by phone or in person. The CPS continues to achieve relatively high response rates, although they have declined from roughly 92 percent a decade ago to roughly 83 percent currently.²

Households participate in the CPS for 4 consecutive months, then drop out for 8 months, and finally return for another 4 months before being removed from the sample.³ To be eligible to participate in the survey, individuals must be 15 years of age or over and not in the Armed Forces. People in institutions, such as prisons, long-term care hospitals, and nursing homes, are ineligible to be interviewed in the CPS. A strength of the CPS is that it collects extensive demographic information for respondents—e.g., on age, gender, education, race, marital status, and family income—which can be used to describe characteristics of different groups for whom monthly labor force data are also collected.

The most recent CWS was conducted as a supplement to the May 2017 CPS. As usual, it asked questions about persons who held a job for pay or profit during the reference week, with a single question also asked of people who did not have a job during that week but looked for work during the last year and were available for work during the reference week. As described in Chapter 1 and as suggested by its name, a central purpose of the CWS is to measure the contingency, or temporary in nature, of the respondent’s main job. The precise questions asked about this differ

___________________

¹ In so doing, it elaborates on the description of the CWS and its role in measuring AWAs in Chapter 1. The description here draws in part on a presentation to the panel at its first meeting (March 29, 2019) by Anne Polivka and Julie Hatch Maxfield and on results from the CWS provided by BLS in an article titled “Electronically Mediated Work: New Questions in the Contingent Worker Supplement” (Current Population Survey Staff, 2018).

² Available: https://www.bls.gov/osmr/response-rates.

³ Available: https://www.census.gov/programs-surveys/cps/technical-documentation/methodology.html.

depending on whether the individual in question is identified in the basic monthly CPS as an employee or as self-employed.

In addition to characterizing whether jobs are temporary, the CWS classifies the respondent’s main job into one of several largely exclusive work arrangements (although they are not entirely exclusive, as on-call workers also can be contract company workers):

• Temporary help agency work
• On-call work
• Day laborer
• Contract company work
• Independent contractor

In principle, independent contractors are by definition self-employed, but the survey’s questions about independent contractor work are asked in the basic monthly CPS both of those who are coded as employees and those who are coded as self-employed (although the questions are worded somewhat differently). For each category, if the respondent (or other household member⁴) is reported to be in the given type of work arrangement, a series of follow-up questions is asked about it. Workers not assigned to one of the alternative work arrangements can be categorized either as employees not in an alternative arrangement or as self-employed but not an independent contractor. Questions were added to the 2017 edition of the supplement to collect information on whether any job (not just the main job) held in the reference period was intermediated by a mobile app or online platform.

After asking about a person’s current job, the CWS asks about the pathway taken into that job. Depending on the person’s job, these can be pathways into independent contractor/self-employment work, into other AWAs, or from AWAs into a current (regular) job. The CWS also includes a set of job satisfaction questions. For the most part, these are cast as “work preferences” and focus on an indicated characteristic of the respondent’s job.⁵ Understanding workers’ motivations is valuable for understanding the implications of AWAs for worker well-being, specifically understanding whether workers are engaged in AWAs as a fallback because standard

___________________

⁴ The CPS allows proxy reports for other members of the household.

⁵ For example, question PES25a, administered to people whose jobs were characterized as contingent, asks, “Would (you/NAME) prefer to have a job that is permanent rather than temporary?” And question PES25aR follows up with: “People have temporary jobs for a variety of reasons. For example, some people have temporary jobs because it is the only type of work they could find. Others have temporary jobs because they enjoy the flexibility or for other personal reasons. What is the MAIN reason (you/NAME) (have/has) a temporary job?” Available: https://www2.census.gov/programs-surveys/cps/techdocs/cpsmay17.pdf.

employer work is unavailable or because the flexible job characteristics of the contract (or other kind of) work are attractive to them.

Finally, the CWS includes detailed CPS-type questions about whether respondents are looking for other work and what has been done to find other employment. It also asks about other characteristics of their jobs, including employee benefits (health insurance, retirement accounts), earnings, hours, and membership in labor unions/associations.

The supplement interview averages 3 minutes, 38 seconds in length for each person who completes it, and 5 minutes, 42 seconds in length for households with at least one person who completes it. An intended 10-minute maximum was exceeded by 12.5 percent of households.

As a supplement to the CPS, the CWS has a number of important strengths. The sample is carefully designed to be representative of the U.S. population, the data collection is undertaken by well-trained interviewers, it has a relatively high survey response rate, and a significant amount of demographic information is obtained about survey participants as part of the basic monthly survey. The fact that the CWS is a survey means that it can be used to collect information about how people feel about their jobs and the reasons for working under particular arrangements, something that cannot be learned from administrative or financial records data. As such, the CWS is well suited to collecting information on important policy-relevant features of jobs and job characteristics, as described in Chapter 2 (section 2.3).

Like any data source, the CWS also has limitations. There are some concerns about the use of household survey data generally, including data from high-quality federal surveys such as the CPS and its supplements. One of these concerns is over the use of proxy reporting. For the CPS, this is a practical necessity, because CPS responses must be obtained during the 10 days following the survey reference week, and it would be difficult if not impossible to interview all household members directly on that schedule. The BLS acknowledges this and notes that, ideally, all respondents would be self-reporters, but given the data collection environment and constraints, eliminating proxy responses simply is not feasible.

Proxy reporters can provide high-quality information about others’ activities and behaviors if they are visible and known to the proxies (Moore, 1988). This may not always be the case, however, with occasional or irregular work. Even when a proxy reporter knows about it, another person’s work activities may be less salient to the proxy reporter than to the person doing the work, so stronger cues may be required to prompt accurate reporting (Abraham and Amaya, 2019). Katz and Krueger (2019) found that in the 1995 and 2005 versions of the CWS, reported participation in AWAs was about 2 percentage points lower among proxy reports than among self-reports. This difference widened to about 2.9 percentage

points lower in 2017. Although these differences could reflect real differences between the people for whom proxy and self-reports were obtained, the fact that they persist even after controlling for respondents’ educational attainment, experience, race, and sex may suggest that proxy reporters have been failing to report some AWA work. All of this suggests that it would be worthwhile for BLS to carry out additional cognitive testing on the ability of proxy respondents to report about AWAs for other members of their households and on the best way to pose questions about this to proxy reporters.

Another factor that may be affecting the accuracy of estimates from the CPS and CPS supplements such as the CWS are the above-noted waning response rates. Falling response rates raise concerns about the representativeness of the CWS sample, because nonresponders are not likely to be randomly distributed throughout the population.⁶ Through 2010, nonresponse rates for the basic CPS were consistently under 10 percent, but that figure has risen substantially in the last 10 years. Even among those who respond to the basic CPS questions, some do not respond to supplements such as the CWS, and supplement nonresponse has grown. In 2005, the overall nonresponse rate to the CWS supplement was 15.5 percent, taking into account both nonresponse to the basic CPS and nonresponse to the additional supplement; in 2017, using the same measure, nonresponse was 23.0 percent.⁷ All of this raises concerns about bias as well as lack of precision in estimates. The CWS response rates are still much higher than those for most other surveys, especially nongovernment surveys, but falling response rates do create a compelling motivation for thinking about ways to complement the CWS with data from other sources.

Issues related to proxy reporting and generally falling household survey response rates across the statistical system are largely beyond the control of the BLS, and they are far from unique to the CWS. Much has been written about both issues in other contexts. In the next section of this chapter, we focus on concerns that are specific to the CWS measurements of AWAs and ways in which those can be addressed. These issues include (1) the short reference period (the prior week), which leads to the strong likelihood that work, especially sporadic work, is missed; (2) the near-exclusive focus on respondents’ main jobs; and (3) uncertainty over whether people can reliably answer some types of questions, such as about the nature of the employment arrangement (e.g., when the employment arrangement is mediated by another party as in the use of temporary help or contract com-

___________________

⁶ The U.S. Census Bureau routinely publishes nonresponse rates, but typically only for a recent 12-month period. Available: https://www.census.gov/programs-surveys/cps/technicaldocumentation/methodology/non-response-rates.html.

⁷ See http://data.nber.org/cps/cpsfeb05.pdf and http://data.nber.org/cps/cpsmay2017.pdf.

pany work). These characteristics of the CWS limit its capacity to measure alternative work in a way that is satisfactorily comprehensive for informing employment policy.

3.2. CWS SCOPE: UNIVERSE OF WORKERS AND WORK ACTIVITIES

Before workers and work can be categorized, job characteristics identified, and trends in work arrangements measured, the universe of respondents (workers) and work activities for which information is requested must be determined. The CWS measures work arrangements concerning the main job of each person who reports having worked during the survey reference period in the basic monthly CPS. By design, it does not capture any work a person may do to generate income that is not reported in response to the core CPS employment questions.

In this section, we consider whether the universe of workers and work activities is appropriately defined in the CWS to capture the information on work in contingent and alternative work arrangements that is needed for policy and research. Specifically, we discuss issues related to the survey reference week, work activities that may not be reported in the main CPS, and secondary work activities.

CWS Universe: Survey Reference Period and Difficult-to-Identify Work Activities

In considering whether the universe of respondents to the CWS is appropriate, two issues deserve discussion. First, we consider whether the survey reference period of 1 week is too short to provide the data needed to understand the role of contingent and alternative work arrangements in Americans’ lives. Second, we consider whether the CWS should allow for the possibility that some work done during the survey reference period is missed in the main CPS—work that is likely to be disproportionately in a contingent or alternative work arrangement.

With regard to the first issue, the designation of the reference period—typically the week that includes the 12th of the month (data collection occurs the following week)—is perhaps the most obvious survey design feature in the CWS affecting the way respondents are sorted by work status. Unlike the work status of people who work a regular weekly schedule or who do not work at all, the work status of a person who, for example, works sporadically throughout the year will depend on the timing of work performed in relation to the reference period.

The issue is that people who report being unemployed or not in the labor force may nonetheless periodically engage in work activities to sup-

plement their household income. Even if respondents always accurately report their work activities during the reference week, this type of periodic work may be missed because of its timing. There is evidence of the potential importance of such supplemental income in both survey data and proprietary sources. The Survey of Household Economics and Decisionmaking (SHED), sponsored by the Federal Reserve Board, asks respondents about their primary work status during the prior month.⁸ All respondents, including those who report “not in the labor force” or “unemployed” as their primary status during the previous month, are asked whether they have worked in any of a variety of side jobs during that month. Estimates from the SHED suggest that holding such side jobs is very common among these two groups: in 2016 and 2017, 20 percent of those who were not in the labor force and 42 percent of those who were unemployed reported having at least some informal work activity or side job during the prior month (Abraham and Houseman, 2019).

Analyses of financial account data indicate that most individuals participating in platform work do so for just 3 months or fewer out of the year, so that the number of platform participants in any particular month is considerably below the number who participate over the course of a year (e.g., Farrell, Greig, and Hamoudi, 2018). Similarly, participation in platform work in any given week is apt to be notably lower than participation over a month or longer period. Again, using financial account data, Farrell, Greig, and Hamoudi (2018) also find a spike in work done for online platforms among those who recently became unemployed, as determined by receipt of unemployment checks. Although such work is often sporadic, it can be an important source of income, particularly for households experiencing financial hardship (Abraham and Houseman, 2019).

Additionally, even if such work is done during the reference week, it may be missed in the CWS. Under the current CWS design, other than a single question asked of unemployed and discouraged workers, the questions on the supplement about current work arrangements are asked only

___________________

⁸ The SHED was administered by GfK, an online consumer research company. In order to create a nationally representative probability-based sample, the survey methodology selects respondents based on both random-digit dialing and address-based sampling. The cumulative response rate for the survey was approximately 4.0 percent for the SHED in 2016, 2017, and 2018. A post-stratification process was used to adjust for any survey nonresponse as well as any noncoverage or under- or over-sampling resulting from the study’s specific sample design. The variables employed in the adjustment of weights for that study comprised gender, age, race/ethnicity, education, Census region, residence in a metropolitan area, household income, and access to the Internet.

of those who are identified as employed in the basic monthly CPS.⁹ Recent research suggests that the basic CPS employment questions may miss certain types of work that are in scope for the survey. Evidence of uncaptured market work done by individuals who may not be categorized as employed in the monthly CPS pertains primarily to what is sometimes termed “side” or “gig” jobs and other informal, nonemployee work activities. In the CWS, this type of work would fall under the category of independent contractor work, broadly defined. Examples of such work could include driving for Uber, tutoring students on the side, doing housekeeping or yard work, providing child care, eldercare, or dog-walking services, or earning income from YouTube video postings.

Robles and McGee (2016), analyzing data from the Enterprising and Informal Work Activities (EIWA) survey fielded by the Federal Reserve Board, report that during the 6 months prior to the survey in 2015, 36 percent of the adult population had participated in informal work that involved either selling or renting property or providing services. Abraham and Amaya (2019) report the results of a study in which a nonrepresentative sample of respondents recruited through Amazon’s Mechanical Turk platform were asked the standard CPS employment questions followed by additional questions about their engagement in various types of independent contractor or nonemployee work. Depending on how the latter questions were phrased, including those who reported such work raised the estimated employment rate by 3 to 5 percentage points. Similarly, results from two waves of the Federal Reserve Bank of Boston Survey of Informal Work Participation (SIWP)—which replicates the basic CPS questions to determine whether a person is employed and also asks separately about participation in informal work activity—show that accounting for informal work raised the employment rate estimated for those age 21 and older from 65.1 percent to 69.6 percent, a 4.5 percentage point increase (Bracha and Burke, 2018).¹⁰

In an analysis of the American Time Use Survey (ATUS), which provides a detailed accounting of individuals’ daily activities, Allard and Polivka (2018) examine the incidence of “income-generating activities” that are not part of any main or secondary job reported. Although they do not find evidence of growth in such activities over time, on average, about

___________________

⁹ The question for those who were unemployed or laid off is about what type of work these people were looking for. Specifically, the survey introduction asks: “I have a question about the type of employment you looked for in the last 4 weeks . . . Were you looking for temporary, short-term employment or more long-term employment?” (with the option of the answer being coded as “either/anything I can find”). There is also a question for discouraged and marginally attached workers, but they do not figure in the weighting scheme and thus a weighted estimate cannot be derived for them.

¹⁰ The SIWP attempts to account for informal paid activity or side jobs, exclusive of selling property, renting property, or responding to surveys.

1 percent of the population reported doing such income-generating activities on a given day from 2012 to 2016. Under the assumption that selected categories of these income-generating activities were in fact “work,” they estimate that correctly classifying as employed individuals who were coded as unemployed or not in the labor force but who engaged in such labor income-generating activities would raise the employment rate for those age 15 and older 0.3 to 1.8 percentage points.¹¹

A possible explanation for these findings from various surveys is that people engaged in nonemployee work or other AWA may think of themselves as unemployed or not in the labor force, especially if they have held a standard employer-provided job in the past. In the existing CWS, however, if such work is not reported in response to the basic CPS employment question respondents are not asked about the arrangements under which it occurs.

The CWS could address these issues—a short reference period and potential underreporting of certain AWAs—by asking CPS respondents who did not report any work in the main CPS a set of initial screener questions.

RECOMMENDATION 3.1: At the beginning of the CWS supplement, screener questions should be asked of those who did not report any work in the basic monthly CPS. The questions should probe into work activities for pay that individuals sometimes do to supplement household income when they are unemployed and looking for a steady job or when they are retired or otherwise not steadily employed. The questions should ask about such work over a longer reference period, such as 1 month, as well as during the CPS reference period (the prior week).

Secondary Work Activities

The scope of the CWS is limited by the fact that it has been concerned only with a respondent’s main job. As discussed in detail in Chapter 2, it would be useful to know about all the work activity of those reporting work, not just about the work associated with one job. Often, secondary work activities are in contract or other AWAs, or they may be held by workers whose main job is in an AWA (Abraham and Houseman, 2019). For various policy-related reasons, it is important to understand when and why people engage in secondary work activities. Particularly important is understanding the circumstances in which people take on second jobs to supplement household income and make ends meet.

CPS data indicate that the share of workers who held more than one job averaged about 5 percent during 2017, the year of the most recent

___________________

¹¹ These results are based on data presented in Allard and Polivka (2018), Table 3.

CWS. In the study described above, Allard and Polivka (2018) also use the ATUS to examine the possibility that the CPS underestimates the prevalence of multiple job holding. Allard and Polivka find evidence of a significant understatement of multiple job holding, concluding that “if workers misclassified as single jobholders were classified correctly, the estimate of multiple jobholders would be between 3.0 percent and 20.7 percent higher in 2012–16 than the current figure.” It is also notable that the base level of multiple job holding in the ATUS is about double that estimated in the basic CPS.

Research based on other sources has generated additional evidence suggesting that secondary work activities are far more common than suggested by the CPS/CWS data and that much of this work takes the form of AWAs. Moreover, the same evidence suggests that income from these activities can be important to households experiencing a reduction in income from a main job, unusually high expenses, or some other type of financial distress.

Data from the above-referenced SIWP produced an estimate of the multiple-job-holding rate that was 11 percentage points higher than the survey’s baseline estimate of 19 percent (Bracha and Burke, 2018). Estimates based on data from the 2016 and 2017 SHED, which included similar questions, were that 28 percent of adults had earned money from informal work or side jobs outside of a main job during the month prior to the survey. Among those adult respondents, roughly one-third, or about 11 percent of the employed, reported that money from their side job was an important source of household income (Abraham and Houseman, 2019). Consistent with those findings, in data collected using a survey module on the nightly Gallup telephone survey, responses to questions designed to capture all work activity, including work involving very low hours, indicated that about 20 percent of workers held multiple jobs (Abraham, Hershbein, and Houseman, 2019).

The results from these surveys suggest that probing for other income-generating work activities substantially increases estimates of multiple job holding. This is a significant finding about the U.S. labor market and, if confirmed in a larger, more representative survey, such as the CWS, is something that would be desirable to have reflected in official economic statistics.

RECOMMENDATION 3.2: For respondents reporting only one job in the basic CPS, the CWS should begin with a set of questions about additional work activity. Respondents would first be asked if they did anything for pay (to supplement income) beyond what they have already reported for their main job. The questions would ask about such work over a longer reference period, such as 1 month, as well as about the CPS reference period (the prior week).

Although the evidence from other surveys showing high rates of secondary work activities is compelling, it is possible that the same result might not hold in the CPS sample. Although they were designed to be representative of the population as a whole, several of the surveys cited earlier were administered through an online panel, and people willing to participate in such panels could be more likely than other people to participate in secondary work activities as well. Nonetheless, cognitive testing for the 2017 CWS uncovered a similar problem. Specifically, the cognitive testing for electronically mediated work indicated that there were multiple paid activities that respondents did not think of as “jobs,” leading the CWS to revise questions for electronically mediated work to include “any work” (Kopp and Edgar, 2016). Thus, the 2017 implementation of the CWS already asked respondents with only one job about additional work for pay using electronic means. The recommendation above simply expands this line of inquiry to include other kinds of work.

Querying respondents about work activity over two time horizons will yield a new estimate of the rate of secondary job holding during the reference week, one that captures second jobs missed in the basic CPS, and an estimate of the rate of secondary job holding over a longer time horizon. As was the case with the employment rate, the estimated rate of secondary job holding over the longer time horizon should be higher than the rate computed in the basic CPS, both because it captures work that is sporadic or interrupted and because the question wording will be designed to better capture work activity that is sometimes missed by standard household survey questions.

A subset (at least) of the work-arrangement questions asked about primary jobs also should be asked about secondary jobs, whether identified in the basic monthly CPS or in response to the new questions about additional work asked on the CWS.

RECOMMENDATION 3.3: It would be desirable to ask the full battery of CWS questions about all secondary jobs held either during the reference week or during the longer 1-month time frame. At a minimum, the CWS should collect information, where applicable, on selected characteristics of one secondary job (when there is more than one secondary job, selecting the one with the most hours worked). These characteristics should include: whether the job is a self-employment or independent contractor arrangement, hours variability, and main reason for holding the secondary job. If not already collected in the basic monthly CPS, information on hours, earnings, industry, and occupation also should be collected.

Cognitive Testing of Screener Questions

Implementing the previous recommendations will require the development of screener questions to identify work activity that may have been missed in the basic CPS interviews. As is always the case when new questions are added to the CPS or its supplements, careful cognitive testing will be required, in this case to determine the optimal reference period and wording to solicit work that was not reported in response to the standard monthly CPS employment questions.

RECOMMENDATION 3.4: Implementation of the new questions proposed for the CWS (in recommendations 3.1 through 3.3) will require extensive cognitive testing to determine the optimal reference period and wording to solicit responses about work that was not reported in the main CPS.

Regarding the reference period, the panel acknowledges that there may be risks in deviating from the standard reference period for a subset of the CWS questions. In testing new questions for the survey, potential risks will have to be weighed against the value of questions about a longer reference period for providing a more complete picture of employment activities. Another concern is that questions about a longer period may be more subject to recall bias. For these reasons, it will be important to pilot the approach recommended above.

Testing can be guided by two hypotheses regarding underreported AWAs. One hypothesis is that AWAs are common but sporadic, so the CWS reports about respondents’ main jobs are correct for the reference week in question. The other hypothesis is that some people engaged in an AWA misreport it (or do not report it). Much of the other work measuring AWAs has used longer reference periods and also has provided respondents with more specific examples of different types of work activity that might be performed under an AWA. This means there are multiple factors that might account for the higher rates of AWA participation in those surveys, making it difficult to directly evaluate the likely impact of asking about a longer reference period. If people are reporting accurately but their work is infrequent or sporadic, then it would be better to ask about the recent week first, and then expand to the 1-month time frame. On the other hand, if there are concerns that revising the survey questions to encourage more complete reporting of AWAs may lead to “telescoping”—unintentionally including in their report activity that occurred prior to the reference period—it will be preferable to start with the longer period and then ask about the shorter period. Asking in this latter order (longer period first, then shorter) puts a boundary around the

answers and reduces the chance that people will misreport activity in the last week when it really occurred earlier.

Aside from the choice of reference period, testing also will need to address how best to obtain complete reporting of primary or secondary work activities not reported in the main CPS. One strategy would be to include examples of the types of work that prior research indicates may be missed in the basic CPS, such as the provision of personal services to households and work obtained online. In addition, it may be worth emphasizing that “even one hour” of work constitutes work for pay. For instance, the American Community Survey asks, “LAST WEEK, did this person do ANY work for pay, even for as little as one hour?”¹² The Gallup survey discussed in Abraham, Hershbein, and Houseman (2019) uses the phrasing, “Thinking about your WORK SITUATION over the past seven days, have you been employed by an employer—even minimally like for an hour or more—from whom you receive money or goods?”

The goal of providing examples is to better convey to respondents the kinds of income-generating work activities about which users of the data would like to know. Following this strategy, the SHED asks about being paid for specific activities, including “child or elder care services; dog walking, feeding pets, or house sitting, house cleaning, yard work, or other property maintenance work” and so on.¹³Jensen, Tickamyer, and Slack (2019) ask about “additional kinds of work—other than the more formal types of employment we’ve already discussed—that many people do to make ends meet” and list 19 different activities. Abraham and Amaya (2019) test questions about informal work that provide detailed examples compared to a simpler question without examples; they find that among those reporting for others in their household, providing detailed examples produced a larger number of reports of that sort of work.

Given budget and time constraints, any examples given in the CWS will need to be more limited than those found in other surveys such as the SHED, which probes participation in about a dozen different work activities. While mentioning some specific company names, such as the names of platform companies, could help to jog respondents’ memories, the relevant company names can be expected to change over time, making that an unattractive strategy for a survey intended to provide information that can be compared over time. For that reason, examples should be framed in terms of the types of work involved (e.g., “ride share work”) and not in terms of specific employers (e.g., Uber or Lyft).

___________________

¹² Available: https://www2.census.gov/programs-surveys/acs/methodology/questionnaires/2019/quest19.pdf.

¹³ Available: https://www.federalreserve.gov/consumerscommunities/files/SHED_2018codebook.pdf.

Finally, while using examples in surveys may improve data quality (e.g., Tourangeau et al., 2014), there are many outstanding issues regarding the selection and presentation of examples in survey questions (see Schaeffer and Dykema, 2020). Therefore, as with the other survey changes proposed in this chapter, the inclusion of examples to help define the constructs under consideration and facilitate recall should be thoroughly tested, because the examples chosen can broaden or narrow what respondents consider in unintended ways.

The BLS naturally may be concerned about asking questions that might affect the CPS responses in subsequent waves. The agency has recognized that some kinds of work may be missed in the basic CPS, and it considered expanding the universe for the questions about electronically mediated employment (EME) arrangements to include unemployed respondents. In the end, the BLS did not modify the survey this way, partly out of concerns that doing so could be seen as challenging people about whether they were really unemployed, potentially affecting reports of unemployment in subsequent waves and causing some disruption to the time series of statistics on employment and unemployment.¹⁴ The BLS also was concerned that asking such questions might appear redundant and frustrate people who thought they already had answered the questions.

There are ways to frame questions about additional work to capture activity that the basic monthly CPS might have missed while mitigating any impacts on subsequent responses to the standard CPS employment questions. For example, in the main CPS a question asked of people who are coded as unemployed or not in the labor force might note that sometimes people do some work on the side to supplement their income (or “while looking for work” if the person is unemployed) and give examples of such work. Similarly, in querying people who report only one job in the main CPS, questions could first note that sometimes people do some work on the side to supplement their income from a primary job, giving examples. Respondents then could be asked if they had done this type of work over the indicated reference period(s).

Particularly for people who report being unemployed or not in the labor force in the main CPS, there is the possibility that the screener questions would capture a primary job held before the individual entered this job status. This possibility could be minimized by wording the question to ask about a job that supplements household income. Further, for those indicating that they held such a job, a follow-up question could directly ask if the job was a primary job that they lost or left prior to becoming unemployed or leaving the labor force.

___________________

¹⁴ “Question universe” section, available: https://www.bls.gov/opub/mlr/2018/article/electronically-mediated-work-new-questions-in-the-contingent-worker-supplement.htm.

Before committing to adding questions about additional work on the CWS, the BLS may wish to test how much additional work activity such a question is likely to uncover. This applies to new questions asked of both those reporting no work and those reporting a single job. The BLS could perform such a test by asking those in the outgoing CPS rotation groups—month-in-sample (MIS) 4 and 8—about work activity not already reported in answering the basic CPS employment questions. Restricting the sample for this test to those in MIS4 and MIS8 would greatly reduce if not eliminate the risk of contaminating later CPS responses while still informing the decision about whether the work activity in question is significant enough to warrant asking about it in the CWS. If additional work activity turned out to be relatively unimportant among CPS respondents, that would create a solid justification for deciding not to ask about it on the CWS. If, on the other hand, consistent with evidence from other sources, a significant number of people report other work activity that the CPS is not currently capturing, that would create a compelling rationale for including questions about that work and the arrangements under which it occurs on the CWS.¹⁵

While the potential to disrupt the historical continuity of the responses to the basic CPS is a legitimate concern, it is critical for policy purposes that BLS endeavor to capture all market work activity. Moreover, as already noted, there are ways to eliminate or minimize any impacts that further probing in the CWS might have on CPS responses in subsequent waves. It is worth disrupting the CWS (and perhaps the CPS) in the ways recommended above, given that the survey addresses profound policy questions about modern employment and how people are piecing together their income. The CWS revision will set a standard for work in this area for a long time to come.

3.3. CATEGORIZING WORK AND WORKERS

All six waves of the CWS measured the contingency of employment—that is, whether a person’s job is temporary in nature—and selected types of AWAs, including independent contractors, temporary agency workers, on-call workers, day laborers, and contract company workers. For all but on-call workers, these AWAs are defined as being mutually exclusive. This section focuses on these mutually exclusive categories, which capture

___________________

¹⁵ One caveat to drawing conclusions based on questions asked in MIS4 and MIS8 is that possible month-in-sample bias may affect reports of employment in the CPS in those waves. For instance, Halpern-Manners and Warren (2012) found that there were significantly lower reports of having a second job among self-reports in the second month-in-sample than there were in the first month-in-sample, for every single first versus second month-in-sample comparison between January 2007 and July 2010. The finding holds even after accounting for changes in mode of interview (see their Figures 6 and 8). If month-in-sample bias is material, then the universe for the CWS questions should be everyone.

instances where a worker is not an employee of the organization for whom he or she is performing work.

The set of AWAs that follows distinguishes between those who are not employees (independent contractors) and those who are in an intermediated arrangement (temporary agency workers and contract company workers). In the latter, workers are employees of a temporary help agency or other type of company that contracts their services to other organizations. Day laborers, hired on a daily or very-short-term basis, most likely are independent contractors. On-demand platform work, captured only in the 2017 CWS, is a hybrid of these two types; while payments to these workers are mediated by the platform company, the workers are usually classified as independent contractors and so may be thought of as a subset of that category.

It is important that a revised CWS continue to collect information on independent contractors, including platform workers, although the panel has some recommendations for modifying the questions that query respondents about these work arrangements. It is also important to collect information on temporary help workers and contract company workers, even though the value of surveying workers about such intermediated arrangements in the CWS is less clear. As discussed in the next section (3.4), rather than attempting to attach a label to on-call and other workers with unpredictable schedules, we recommend collecting information on these job characteristics, which would allow data users to categorize work arrangements as appropriate for their needs.

Independent Contractors

As discussed in Chapter 2, there is strong policy and research interest in distinguishing between those who are W-2 employees and those in various self-employment or nonemployee arrangements. The latter are not covered by employment and labor laws, such as wage and hours laws and laws giving workers the right to form a union and collectively bargain. They generally are not covered by social insurance programs, such as unemployment insurance and workers’ compensation. Finally, they do not have access to benefits that many employers provide their employees, such as paid sick leave, health insurance, and retirement benefits. The large number of people in nonemployee work arrangements and the possible growth in these arrangements’ prevalence has raised concerns about the adequacy of social protections in the United States. It is therefore critical that the CWS try to distinguish between those who are W-2 employees and those in independent contractor and other self-employment arrangements.

Among those identified as self-employed, the basic CPS differentiates only between the incorporated and the unincorporated self-employed. The

CWS also differentiates between those who are independent contractors, independent consultants, or freelance workers and other self-employed individuals. A challenge in household surveys in measuring self-employment and subcategories of self-employment such as independent contractors is that respondents may have quite different understandings of what these terms mean. For example, while some respondents may understand that self-employment means that they are not employees of an organization, others may understand self-employment to mean that they own and operate their own business. Those who work for an organization but are not a W-2 employee may not think of themselves as self-employed, and in common parlance may still refer to themselves as employees. Similarly, usage of the term “independent contractor” may carry different connotations and vary across occupations. In some occupations, for instance, distinctions are made among nonemployees between independent contractors and subcontractors. Idiosyncratic differences in the use and understanding of such terms may significantly affect how people respond to questions in household surveys about their independent contractor status.¹⁶

In the current CWS, respondents who identified both as self-employed and as employees in the main CPS are asked whether they are an “independent contractor, independent consultant, or freelance worker” on their main job. Those who identify as working for an organization in the main CPS (and hence are coded as employees) are also told that independent contractors, independent consultants, and freelance workers obtain customers on their own to provide a product or service. Although independent contractors should not be identified as employees in the CPS, recent research evidence provides compelling evidence that many likely are, which is consistent with the widespread confusion over these terms. Whereas the share in self-employment has not grown in the CPS, research using administrative data shows a sizable growth in the share of individuals with self-employment income as reported to the tax authorities and higher levels of self-employment than found in the CPS (Abraham et al., 2020; Abraham et al., forthcoming; Jackson, Looney, and Ramnath, 2017; Katz and Krueger, 2019; Lim et al., 2019). Some of this reflects work activity that is not mentioned in household survey responses, but some of it reflects self-employment work activity that is miscoded as wage and salary work.

One potential reason for the undercount of self-employment in the CPS is that the survey may be prone to misidentifying independent contractors as employees (Abraham, Hershbein, and Houseman, 2019). In the CPS, individuals who indicate that they did work for pay are asked: “Were you

___________________

¹⁶Abraham, Hershbein, and Houseman (2019) provide a thorough discussion of definitional issues in measuring independent contracting.

employed by government, by a private company, a nonprofit organization, or were you self-employed or [if applicable] working in the family business?” Individuals responding that they are employed by the government, a private company, or a nonprofit organization are coded as employees. It would be reasonable and accurate for a respondent working on a contract basis for a company or organization to report being employed by that entity, particularly if the term self-employment carries certain connotations, such as running one’s own business.

In a Gallup survey module on contract work, Abraham, Hershbein, and Houseman (2019) found that a sizable share (about 10%) of Gallup respondents who reported being employed by an employer, and so were coded as employees, also indicated upon further probing that they were an independent contractor, independent consultant, or freelance worker. Linking tax records to data from the Annual Social and Economic (ASEC) supplement to the CPS for the period from 1996 through 2015, Abraham et al. (2020) and Abraham et al. (forthcoming) find substantial growth in self-employment in the tax data that is not reported in the CPS-ASEC. Their findings are similar to those from other research using administrative data. Consistent with the findings reported by Abraham, Hershbein, and Houseman (2019), about a third of this growth was accounted for by people who reported only employee work in the CPS-ASEC and only self-employment work in the tax data.

Given this evidence, a broad approach is needed for measuring independent contract work. Questions pertaining to independent contractor work in the CWS should be designed to capture individuals who (a) are not employees of an organization and (b) do not own or operate a business or have significant capital investment in their business. Currently, no definition of independent contractors, who are also commonly referred to as consultants or freelance workers, is provided for those identified as self-employed in the CWS. Many of those who are in informal work arrangements—such as providing child care, elder care, cleaning, or maintenance services directly to households or providing services through online platforms or mobile apps—also should be captured in the independent contractor category.

Moreover, the definition of independent contractors given to those identified as wage and salary workers in the CWS—“someone who obtains customers on their own to provide a product or service”—may be misleading because this is not a defining characteristic of independent contractor work. Many who work as independent contractors do so primarily for one organization and are commonly termed “dependent” contractors to reflect that reliance on a single client. Individuals in this arrangement might not think of themselves as obtaining customers on their own in the same way as, say, a self-employed business owner might. Those who report being employed by an organization in the main CPS but say in the CWS that they

are independent contractors—between 1 and 2 percent of those identified as an employee in the main CPS—may be especially likely to work primarily for a single client. This may mean that the CWS data understate the share of CPS employees who are in fact independent contractors. We recommend removing the definition of independent contractors as people who obtain customers on their own from the question.

RECOMMENDATION 3.5: The CWS should continue to ask those identified both as self-employed and as employees in the main CPS about their status as independent contractors. A broad definition of independent contractor should be given in both questions, and the current definition of independent contractor used for those identified as wage and salary workers should be replaced.

For those who are identified as self-employed in the main CPS, the goal of the question is to identify those who do not own or operate a business or who do not have a significant capital investment in their business. The goal of the question asked of those identified as wage and salary workers in the main CPS is to distinguish between W-2 employees and nonemployees; the measure for independent contractor work should include the types of informal work activities captured in the SHED and other recent surveys in which the individual is not an employee.

RECOMMENDATION 3.6: Cognitive testing should, among other things, determine how well respondents distinguish between employee and nonemployee concepts and explore ways to improve the accuracy of responses. To this end, BLS might clarify for respondents that independent contractors are not employees of the organization or customers for whom they provide a good or service and/or do not have any taxes taken from their pay.

The first step in testing is for BLS to evaluate how respondents understand the CWS questions about independent contracting, especially for proxy reports. Errors may arise, either because respondents do not understand the questions as written or because the nature of independent contracting or freelance work has changed over time and respondents cannot easily align their work into these categories. Even someone who has reported their work activity in the basic monthly CPS—someone who sells jewelry on Etsy and at the local flea market, for example—might not think of him- or herself as an “independent contractor” or “freelance worker.” As noted above, people who work on a nonemployee basis for a single firm may be an especially problematic group. Another possible source of error is interviewers who do not administer a question as intended. Initial work

on the administration of these questions in the 1995 CWS indicated that interviewers often changed the question wording during administration, especially in households with multiple adults (Polivka, 1996). All of these potential issues should be evaluated.

For those identified as wage and salary workers in the main CPS, one option, tested by Abraham, Hershbein, and Houseman (2019), is to ask respondents if their employer withheld taxes from their pay. At a minimum, payroll taxes should be withheld for everyone who is a wage and salary worker, but firms do not withhold taxes from payments to independent contractors. BLS could perform cognitive testing on the approach used by Abraham, Hershbein, and Houseman as one way of distinguishing between employee and nonemployee work.

Platform Workers (subset of independent contractors)

Work mediated by web platforms still represents a small percentage of independent contracting, but its rapid growth has raised the profile of policy issues that more generally concern independent contractors. As discussed in Chapter 2, some people engage in web-platform work to supplement their income, but others rely on it as their main source of income. This raises concerns about workers’ ability to access basic benefits such as medical care and retirement savings. Additionally, Internet intermediary companies have formalized many work arrangements previously considered informal, and there is some evidence that web- mediated options are displacing more traditional job arrangements. One example is Jitjatjo, a “human powered contingent labor platform and staffing marketplace” that seeks to intermediate between businesses in the service and hospitality industry (primarily restaurants) needing to hire on short notice and stopgap workers willing to step in.¹⁷ Another example is Pared, an app used by restaurants (including many high-end establishments) to make on-demand hires of chefs and other culinary professionals.¹⁸

Such developments have spillover effects on wages in employer-provided jobs. So, even if these platform companies account for only a small share of jobs, these new ways of engaging with workers can lead to outsized impacts in the broader labor market. For these and other reasons, BLS has prioritized improved measurement of web-platform work. As noted earlier, new questions introduced in the 2017 CWS were designed to capture work obtained through platforms and mobile apps where payment was coordinated by these

___________________

¹⁷ Available: https://www.jitjatjo.com/index.

¹⁸ Available: https://www.washingtonpost.com/lifestyle/food/apps-have-turned-restaurantwork-into-a-gig-economy-hustle-heres-how-one-cook-chases-a-paycheck/2020/02/24/1f02ee5c-54a8-11ea-9e47-59804be1dcfb_story.html.

companies. However, the new questions did not work as intended and resulted in unrealistically high estimates of the number engaged in this type of work. Following a detailed examination of the data, the BLS recoded many of the answers to these questions that were deemed likely to have been wrong.

RECOMMENDATION 3.7: The CWS should continue to ask about work mediated by platforms/apps. These questions will require extensive modification and further testing.

Additionally, a decision must be made about what kinds of web-mediated activity to consider within the scope of the CWS. One key distinction that is often drawn is that between “work-based” income and income derived from a combination of work and capital. This distinction is somewhat blurry, however, as many web-mediated jobs combine capital and labor inputs. An illustrative example, described in Chapter 2, is renting out rooms through Airbnb, which entails both labor and nonlabor inputs. Although renting out an Airbnb home is typically thought of as a capital-based income-generating activity, a person operating several properties may spend substantial time in the enterprise. How far apart Airbnb and Uber are on the labor/capital continuum is an open question.

RECOMMENDATION 3.8: For purposes of measuring web-platform work, BLS should test the option of not making a capital/labor distinction. The survey could simply ask self-identified platform workers which company (or companies) they work with, and then allow the data to be sorted depending on the question at hand. If the concern is over what motivates people’s efforts to generate income, the distinction regarding the extent to which income is a return on labor or capital may not be crucial.

Another point is that respondents may have difficulty differentiating between platform work that is capital-based and that which is labor-based, or where on the spectrum their activity falls. An advantage of simply including all income-generating activities, whether they are slightly more labor-based (Uber) or slightly more capital-based (Airbnb), is that doing so avoids requiring respondents to make this difficult, subjective call.

Another distinction that may be important to make in order to meet future measurement needs is between work that is provider/client-based and work that is mediated through another company. In some cases, the two arrangements may involve an almost identical type of work. Consider, for example, a person who walks and boards dogs through his or her own neighborhood business, dealing directly with customers, while another person provides the same services but does so through the Rover.com app.

Again, the distinction is between serving a customer (final demand) and working through an intermediary business such as a platform company. The CWS should aim to provide data to help researchers and policy makers understand where—in what sectors, locations, and so on—platform work is expanding and how this work may be changing labor market patterns.

Finally, given the difficulty that respondents had in understanding questions pertaining to platform work on the 2017 CWS, these questions should undergo much cognitive testing. Having the name of the platform company will potentially be valuable for conducting a quality check on data from these questions in a future survey.

Temporary Help Agency Employment and Contract Company Work

Some questions in the current CWS have proven less useful than others, either because evidence indicates the quality of the resultant data is poor or because for other reasons the data are little used. Such cases present an opportunity to create survey space by streamlining parts of the survey. The temporary help agency component of the survey is one such candidate. While tracking temporary help agency employment is important, the incidence measured in the CWS is well below that indicated by employer surveys and administrative data (Polivka, 1996). Respondents appear to confuse their employer, which is the temporary help agency, with the client for whom they are performing work.

Another candidate for CWS streamlining pertains to information on contract company workers—those who perform work for a client on or off the client’s worksite. Although information about contract work is needed, household surveys may not be the best vehicle for collecting it. Partly because BLS determined that respondents cannot reliably report on the contract arrangements of their employer, the CWS focuses on measuring a narrow subset of contract company work: individuals who work primarily for one client company at the client’s worksite. This narrow definition misses a lot of contract work that is relevant for policy, such as contract workers who work remotely. Further, BLS reports that its contract company worker measure is little used. For these reasons, measuring this category of AWA is not a comparative advantage of the CWS.

RECOMMENDATION 3.9: BLS should consider dropping questions on temporary help agency employment and on contract company work from the CWS to make room for other, higher-priority questions.

If the BLS opts to continue to ask about temporary help agency employment in a future CWS, it should consider revising the question wording to use terms like staffing agency in lieu of, or in addition to, temporary help

agency. The industry has long used the terms staffing firm and staffing agency instead of temporary help agency; use of the older term may contribute to the undercount of these workers in the CWS.

Data supplied by businesses are largely complementary to those captured in household surveys and can fill in some information needs. Establishment- and firm-level surveys are an underexplored source of data on alternative work arrangements, especially subcontracted work, that have the potential to yield vital information about the prevalence and nature of firms’ contracting-out activities.

On-call Workers and Day Laborers

The current version of the CWS asks respondents if they are on-call workers. A key attribute of both on-call and day laborer work is that hours are unpredictable—the person works only when he or she is called in or someone offers them work. The predictability of hours is an important dimension of a person’s work activity that the CWS should ask about. As described in the next section, however, the panel recommends that this be done by asking directly about work scheduling and the predictability of hours, within the context of inquiring more generally about the characteristic of the person’s job or jobs, as opposed to asking whether the person identifies as an on-call worker or day laborer.¹⁹ Depending on the arrangement, a day laborer also may be identified through questions pertaining to independent contractors or temporary help workers.

3.4. INSECURITY IN HOURS, JOBS, AND EARNINGS

Capturing the security of workers’ jobs has been a primary focus of the CWS over its history. The concept of contingency in the CWS measures whether a worker’s job is temporary because the position is expected to last for a limited time. This concept of contingency captures job insecurity that, in cases where loss of the job results in employment gaps, may lead to

___________________

¹⁹ Early testing of the CWS indicated that the questions for on-call workers, contract workers, and independent contractors/consultants/free-lance workers were “too long,” resulting in a notable number of interruptions and, in some cases, failure of interviewers to read definitions to the respondents (BLS, 1995). In the current questionnaire, the information defining these job categories is in parentheses, presumably indicating that it is not to be read to persons in the household beyond the first respondent. This questionnaire structure is notably difficult for interviewers to deal with, however, as each one must decide how to implement the parenthetical information. This may lead to increased interviewer-driven variation in answers and potentially different distributions of answers for those who are exposed versus not exposed to information in the parenthetical statements (Dykema et al., 2016; Olson, Smyth, and Cochran, 2018).

employment and earnings instability. Job insecurity is one factor that may result in earnings insecurity.

Even if the job itself is not temporary, the hours and hence a worker’s earnings may be variable. In addition to on-call work, other types of scheduling practices also may be associated with unpredictable hours and earnings insecurity. While the alternative work arrangements described in the preceding section are mutually exclusive, job contingency and unpredictable schedules are characteristics of jobs that may be present in all or most work arrangements, including when workers are employees of the organization for which they perform work.

Schedule Variability and Predictability

As noted above, the CWS asks wage and salary workers whether they work on an on-call basis or whether they work as day laborers. These forms of on-demand work are associated with variable hours and unpredictable schedules that may result in earnings insecurity.

RECOMMENDATION 3.10: As opposed to the current approach of classifying workers into the categories of on-call workers or day laborers, the CWS should focus on simply describing the characteristics of these and other work arrangements with variable hours.

Subsequently, analysts can sort workers by scheduling arrangement, but respondents do not need to be asked directly to label the category into which their employment falls. This job-characteristics-based approach will mitigate confusion over terminology and hence potential misreporting on the survey. For example, an individual may not think of herself as being an “on-call worker” even if she must report to work when called.

While the current version of the CWS only asks about hours variability and uncertainty as they pertain to on-call and day laborer work, because they represent an issue of growing importance variability in hours and scheduling uncertainty need to be measured more comprehensively. As noted in Chapter 2, the widespread adoption of scheduling algorithms in the retail sector and other services industries has meant that firms are better able to match workers’ hours to the company’s needs. The resulting variability and unpredictability in work hours shifts scheduling and income risk from the firm onto the workers. Even if the total number of hours a worker is offered remains constant, if a worker’s schedule is so erratic that she does not know when she will have to come in to work that in itself can take a toll—for example, by making it difficult to arrange child care.

Research on work scheduling issues²⁰ provides ample justification for adding question about scheduling in the CWS. National surveys are typically limited in capturing this dimension of work; since question wording is often about “usual or typical hours,” by design, the resultant data smooth rather than reveal variations in work hours.

By contrast, some specialized surveys have allowed researchers to estimate the magnitude and direction of fluctuations in weekly work hours. For example, looking at scheduling variability in the National Longitudinal Survey, Lambert and colleagues (2014) find that 83 percent of respondents age 26 to 32 who are hourly part-time workers give a different number for the greatest and fewest hours they worked in a week. The average difference between those two is 13 hours, which amounts to 37 percent of their reported usual hours. For these estimates, the researchers developed a survey, the Work Scheduling Study, which asks: “In the last month (past three months), what is the greatest number of hours you’ve worked in a week at this job? Please consider all hours, including any extra hours, overtime, work you did at home, and so forth” and “In the last month (past three months), what is the fewest number of hours you’ve worked in a week at this job? . . . Please consider all hours, including any extra hours, overtime, work you did at home for your job, and time you spent on work that may not have been directly billable or compensated.” These analyses provide models of how to move forward on measuring work schedule variability.²¹

High schedule variability often brings with it a lack of predictability, which can create additional problems for a household and should therefore be tracked. Using data from the SHED and the 2016 General Social Survey (GSS),²²Fugiel and Lambert (2019) find that the number of workers reporting that they receive less than 7 days’ notice regarding their work schedule is 15 percent. For those receiving less than a day’s notice, the figures for the two surveys are 6.4 percent (GSS) and 5.1 percent (SHED).

Although surveys such as the SHED and the GSS have measured hours or schedule variability, they rely on small samples. Given the prevalence of erratic and unpredictable worker schedules, this aspect of employment and earnings instability should be measured with a much larger, nationally representative survey.

___________________

²⁰ See, for example, Henley and Lambert (2014) and Schneider and Harknett (2019). Another article by Miller (2019), “How Unpredictable Work Hours Turn Families Upside Down,” provides case studies.

²¹Lambert et al. (2019) provides a full set of statistics on the prevalence of schedule instability, unpredictability and lack of schedule control using data from the General Social Survey.

²² Available: https://gss.norc.org/.

In addition to questions attempting to get at workers’ lack of control over when and how much they work, it may also be worth asking about more favorable work arrangements, such as whether the worker is allowed flexibility in his or her hours.

As stated in Recommendation 3-10, with this kind of detailed information about workers’ schedules it should not be necessary to ask respondents to self-identify their work category. Appropriately identified characteristics capture the aspects of being, say, an on-call worker that are most salient to workers’ well-being and to the policies designed to safeguard it. Likewise, “on-call” workers may have regular work, or they may not be called in to work very often or very predictably. Some workers, for example those in retail, may be called in to work and then given no hours. The key policy question is, “which (primary) jobs are providing insufficient income, or highly unpredictable income?” Including a set of scheduling questions (restricted to people working for employers)²³ focused on hours and earnings variability at the core of the new CWS would get at the kind of variability that is likely a common motivation for pursuing a second job.²⁴

A number of surveys offer options for question wording related to characteristics of jobs rather than labels of AWAs. For example, the 2017 American Time Use Survey Annual Leave Module asked respondents about the availability of a flexible work schedule, since sometimes an inconsistent—though probably not unpredictable—work schedule is what a worker

___________________

²³ Ideally, in order to generate information on how schedule instability and unpredictability vary across today’s employment arrangements, one would like to measure scheduling patterns of independent workers as well. Asking nonemployee workers about scheduling and hours, however, would require another battery of questions—an expansion of the CWS that, while desirable, may not be feasible.

²⁴ Asking about variability in hours (and not earnings) may be sufficient for the wage and salary group.

needs or prefers. The module also asked about the reasons for being in a flexible work schedule, about days usually worked (including an “it varies” answer), and about working from home. Cognitive testing revealed issues with respondents’ answering that yielded changes in the question wording for the implemented module (Mockovak and Kaplan, 2015).

Contingency

The current CWS focuses on measuring contingency—that is, whether the job is temporary in nature. Indeed, concerns that temporary, or contingent, jobs were a growing share of employment motivated the initial development of the CWS in the 1990s. A high percentage of questions in the current version of the CWS are devoted to measuring the contingency of work, and the BLS publishes three measures of contingency based on the survey results. In the May 2017 CWS, BLS reported that 5.9 million people, or 3.8 percent of the employed, held contingent jobs, according to the broadest measure of contingency.²⁵

Over its history, the CWS has uncovered no growth trend in contingent jobs. Instead, there has been a small but steady decline over time in the estimated percentage of the employed holding contingent jobs, and BLS’s measures of contingency have not gained currency among policy makers and researchers (see, for example, the wide-ranging definitions of contingency used in U.S. Government Accountability Office, 2015). Moreover, the BLS definitions sometimes relied on people’s expectations about future employment, which BLS’s own cognitive testing found to be problematic. Respondents struggled to answer how long they expected their job (or the job held by another household member) to last, which resulted in data that were likely of low quality. Quality assessment research on the CWS indicated that respondents did not know how to answer the questions, “How much longer do you expect to work in your current job/to be self-employed?” and “Do you think it will be more than a year?,” resulting in nonspecific answers, such as, “Well, I don’t know” or “It depends on the economy and GM’s future hiring plans” (BLS, 1995, p. 5). In addition, there was notable interviewer variation in addressing these difficulties. Although there are options to select “something else,” including “Until I retire” or “As long as I want,” interviewers did not appear to select these to summarize respondents’ answers very often.

In the U.S. economy, many jobs are temporary in some sense. The country has an “employment at will” doctrine such that, except for certain reasons and protected classes, workers can be fired by their employer at any

___________________

²⁵ Available: https://www.bls.gov/opub/ted/2018/3-point-8-percent-of-workers-were-contingentin-may-2017.htm?view_full.

time for any reason. Unlike many other economically advanced countries, the United States does not have a strong legal system of employment protections. Key exceptions are that employers cannot fire workers on the basis of gender, religion, ethnicity or race, or age. In this employment environment, many workers might view their jobs as temporary.

Seasonal work may or may not be considered temporary by respondents, especially if they return to the job each season. Similarly, many people work from project to project and, while projects end, people may be reasonably assured of a pipeline of work going forward. It is therefore unclear (including to respondents) what the concept of temporary means in the modern labor market. These concerns argue for deprioritizing the contingency questions in the CWS.

RECOMMENDATION 3.12: While the temporary nature of some jobs is a key characteristic that should continue to be measured in future CWS surveys, the number of questions on contingency should be pared back and the questions that remain should be simplified.

In summary, the panel recommends against classifying workers as on-call workers or day laborers and instead recommends expanding questioning about job characteristics to include a broader set of arrangements associated with unpredictable work schedules. Some questions pertaining to contingent work should be maintained, although the panel recommends paring and simplifying these questions.

3.5. OTHER INFORMATION NEEDED FOR UNDERSTANDING ALTERNATIVE WORK ARRANGEMENTS AND THEIR IMPLICATIONS FOR WORKERS

Key goals of the CWS involve not only measuring the number of people in contingent and alternative work arrangements but also facilitating an understanding of the relationship between specific work arrangements, worker characteristics, and worker outcomes. Because the CWS is a supplement to the CPS, detailed information about the demographic characteristics of respondents is available. The CWS adds measures of key worker outcomes that are not collected in the main CPS: earnings²⁶ and benefits as well as indicators of individuals’ preference for or satisfaction with their work arrangements. The CWS also collects information on workers’ job histories and their transitions into and out of selected work arrangements.

___________________

²⁶ The main CPS collects information on earnings for the outgoing rotation groups (MIS4 and MIS8). The CWS collects information on earnings for other individuals, primarily for those who indicate being in a contingent or alternative work arrangement.

Wages and Benefits

The linkage between people’s income and their well-being is clear; for this reason, it is essential to include questions about earnings in the CPS/CWS. As discussed in Chapter 2, researchers and policy makers have been keen to understand how changing work arrangements, including the expansion of AWAs, may relate to the flat earnings growth experienced by workers over recent decades. The CWS collects information on earnings from those identified as being in a contingent or alternative work arrangement and selected other workers. These data may be supplemented with earnings data collected in the main CPS for those in the outgoing rotation groups (MIS4 and MIS8). So that the earnings of those in alternative work arrangements may be compared to those in traditional arrangements, it is important that the sample for the latter be representative of that population; the BLS should provide weights appropriate for use in analyses of earnings, including analyses that compare the earnings of those in alternative work arrangements with the earnings of other workers.

The interaction of an employer-delivered worker benefits system and emerging AWAs with limited or no access to employer-provided retirement plans, medical care plans, and other benefits is also of great concern. The possible link between contingent and nonstandard work, wages, and access to critical benefits such as health insurance and retirement benefits is of strong policy interest. The onset of a global pandemic has underscored the need for policy makers to be more aware of the economic vulnerabilities present in the labor market. The CWS will continue to shed light on these trends since it includes questions that ask about wages and benefits. Additional information will be obtained by asking about the wages and hours associated with respondents’ second jobs (as proposed in recommendations 3.2 and 3.3).

Capturing net earnings—gross revenues less costs—is important for independent contractors. In a study of earnings from driving for Uber and Lyft in the Denver area, for example, Henao and Marshall (2019) find that, whereas gross hourly earnings averaged $15.87, earnings less costs ranged from $5.72 to $10.46, and most drivers earned less than the minimum wage. For independent contractors, asking about earnings on an annual basis may be most practical, because expenses tend to accrue unevenly over the course of the year. Questions about earnings posed to independent contractors should clearly instruct them to report net rather than gross earnings, to the best of their ability.

Job Satisfaction and Reasons for Choosing a Work Arrangement

The CWS includes a set of questions about workers’ preferences for specific job characteristics in Section 2, “Worker Satisfaction with their

Current Employment Arrangement.” The questionnaire is structured such that workers with particular arrangements—temporary help agency workers, on-call workers, day laborers, and self-employed/independent contractors—are first asked a tailored question about whether they would prefer a job with a different arrangement (e.g., on-call workers are asked if they would prefer a job with regularly scheduled hours). Next, workers are asked for their main and secondary reasons (if their main reason is “for the money”) for working in their current job arrangement. Temporary workers are also asked if they would prefer a job where they could stay as long as they want. These questions for different categories are summarized in Table 3-1.

Measuring the degree to which workers are satisfied with AWAs is important for understanding the extent to which people take alternative jobs by choice or out of necessity. It is unclear, however, how the current set of job preference questions should be interpreted. It can be difficult for respondents to specify preferences unless a set of tradeoffs is provided. For example, if a survey interviewer asked a person if they “would prefer to not work most nights and weekends?”, many would reply, “Yes.” But it also may be the case that most of these respondents would not want to change careers, so the more informative answer is no, given the tradeoffs (Mas and Pallais, 2017). In the context of the CWS, most respondents who are in on-call arrangements report that they would prefer regularly scheduled hours. Yet, these individuals may nonetheless prefer this arrangement if, for example, it pays a relatively high wage.

Moreover, the wording of the preference questions varies across contingent and alternative work arrangements, and job preference questions are not asked of those in traditional wage and salary arrangements. To provide a basis for comparison, it is desirable to use the same question wording to measure how workers across various work arrangements view their jobs. Instead of asking the current set of job preference questions for workers in selected arrangements, the CWS could ask all respondents a question about their overall satisfaction with their job. For example, the General Social Survey’s Quality of Working Life Module asks, “All things considered, how satisfied are you with your (main) job?”²⁷

The existing line of questions on the reasons individuals work in alternative arrangements provides potentially valuable information, but the fact that they are “field coded” makes these questions problematic as well. For a field-coded question, the respondent is asked to provide a narrative response that the interviewer codes using a set of predefined, categorical options. A number of studies indicate interviewers vary in their ability to

___________________

²⁷ Available: http://gss.norc.org/Documents/codebook/QWL%20Codebook.pdf.

code these types of responses accurately (see review in West and Blom, 2016). For example, Smyth and Olson (2020) find that in one national telephone survey, interviewers accurately coded only 49 percent of answers into nominal categories. The challenge of accurately classifying respondents’ narrative responses is likely compounded when the categories are not comprehensive (when they do not include the range of possible categories) or when the labels for the categories are difficult for interviewers to use. Cognitive testing of field-coded questions can help ensure that the labels for the categories included in the questionnaire accurately reflect the range of likely responses and use appropriate language. An alternative to field coding the questions is to narrow the options to a short list of possible reasons and ask the respondent to select the most important.

As discussed throughout this report, it is also important to know why people are engaging in multiple jobs—including why they are doing web-mediated work on the side. Questions on job satisfaction and on the motivation of individuals to work a second job/work activity would be extremely useful for understanding people’s work patterns.

RECOMMENDATION 3.13: BLS should ask questions on job satisfaction for all workers in lieu of asking job preference questions for selected workers. The BLS should continue to ask about reasons for working in selected alternative work arrangements, but it should consider moving away from field coding the responses and instead provide a preset list of reasons and ask respondents to identify the most important. In addition, for those with a second job or work activity, BLS should ask about their motivations for holding multiple jobs.

Job History and Work Transitions

The CWS asks a series of questions about people’s history in their main job. The primary item collected concerns job tenure, with the question wording differing according to the reported employment arrangement. Information on job tenure can be useful for some research and policy purposes. If retained in a future CWS, the structure of the questions could be simplified to ask about tenure on the primary and, if applicable, secondary job, regardless of whether the job is in self-employment or as an employee.

One could make an argument for dropping the job tenure questions. Early cognitive work on the CWS indicated that the questions about how long an individual has worked for their current or former employer yielded highly imprecise answers from respondents: More than 35 percent of respondents answered with a qualified answer, or with an answer that did not match the question task (inadequate), or else said they did not know. Most of these imprecise answers were given when this retrospective question was proxy-reported (discussed further below).²⁸ On balance, however, even if respondents find it difficult to give a precise answer to the job tenure question, the responses are likely to contain valuable information.

Asking about other aspects of respondents’ job histories, however, seems both more problematic and less likely to be valuable. The current version of the CWS asks those who report working for an employer whether they previously worked for the employer under a different work arrangement. Although this information potentially provides interesting background, the question is retrospective in nature and subject to recall bias; data quality is likely to be particularly poor for proxies. For this reason, and because the information collected in prior years has been rarely used, these questions are candidates for dropping in future iterations of the CWS.

Another set of questions in the CWS asks respondents about transitions into their current employment arrangement; specifically, what they or other household members were doing prior to becoming an independent contrac-

___________________

²⁸ Quality Assessment Research on the CWS, 1995, p. 8.

tor, a temporary agency worker, or an on-call worker or contract company worker. These questions also are likely to be subject to recall bias and have been little used, and thus are candidates for dropping as well.

An additional section of the CWS asks respondents in AWAs or contingent/temporary jobs if they have looked for other types of work in the last couple of months (since the beginning of December for the CWSs conducted in February and since the beginning of March for the May 2017 CWS). If they answer yes, respondents are asked for details on their job search. Searching for another type of job may be a good indicator that workers in alternative or temporary arrangements are dissatisfied with the job. The line of questioning in this section, however, is not a high priority, and BLS should consider dropping these questions as well. Also, since only the people in AWAs are asked these questions, there is no benchmark against which to compare their responses. One could imagine that it would be useful to know whether the share of people engaged in on-the-job job searching is higher among those in AWAs than among those in other types of jobs. Perhaps this could be asked without then going into detail about exactly how they had searched.

RECOMMENDATION 3.14: BLS should consider dropping questions on respondents’ job history (except, possibly, those concerning job tenure), on their transitions into their current employment arrangements, and on whether they have looked for employment in another type of job.

Researchers have effectively exploited the longitudinal structure of the CPS to study transitions into and out of the contingent and alternative work arrangements (Addison and Surfield, 2006, 2009; Farber, 1999; Houseman and Polivka, 2000; Katz and Krueger, 2017). In general, as discussed in Chapter 4, there are panel datasets that are better at characterizing people’s job histories. Also, every 2 years, BLS fields a Job Tenure Supplement of the CPS (also known as the Employee Tenure and Occupational Mobility Supplement) that includes questions about respondents’ type of work, industry, and occupation from 1 year earlier and on how long respondents have worked for their current employer.²⁹ This supplement, therefore, duplicates coverage on some aspects of job history. Hyatt and Spletzer (2016) use these data to show the distribution of jobs in terms of duration held. Likewise, the CPS Displaced Workers and Job Tenure Supplement may provide a home for some question areas. The stated goal of that supplement is to “measure the severity of job displacements and assess employment stability

___________________

²⁹ Available: https://cps.ipums.org/cps/jt_sample_notes.shtml.

during a period of downsizing at many firms, and increased use of temporary and contract labor.”³⁰

In summary, in addition to basic information on wages and benefits, BLS should ask CWS respondents a question on job satisfaction. The question should be asked of all workers, including those who do not fall into any alternative work, contingent work, or unpredictable scheduling arrangement. Understanding the degree to which workers are satisfied with (or prefer) their arrangement is important, but a comparison is needed to assess whether workers in various alternative arrangements are more or less satisfied than those in regular employment. Other surveys could be consulted to guide question wording. Questions about main and secondary reasons for having a job would be retained but revised as appropriate for the work characteristic. Questions on respondents’ job history and employment transitions could be eliminated or cut back to create survey space for the new questions.

___________________

³⁰ Available: https://catalog.data.gov/dataset/current-population-survey-displaced-workersjobtenure-supplement.