Working hard or hardly working

On work (over)estimation

Source

There are two main ways of assessing how much someone works — asking them (“how many hours did you work?” and taking a time diary and adding up all the reported work hours). Sometimes these two measures differ, and time diary estimates are usually considered to be a more accurate survey measurement.¹

On an individual basis, these differences are not necessarily meaningful. The American Time Use Survey, and most large American time use surveys prior to 2003, uses a single-day time diary. A respondent could usually work 8 hours a day Monday through Friday, but fill out the time diary for a Saturday and report zero hours of work. However, over a large enough sample, and with proper weighting, time use researchers are able to see how time diary work estimates and survey question work estimates differ.

Overestimating the Work Week

In chapter 2 of Changing Rhythms of American Family Life (first post) Suzanne Bianchi, John Robinson, and Melissa Milkie cited “The Overestimated Workweek? What Time Diary Measures Suggest?” by John Robinson and Ann Bostrom in 1994. Robinson and Bostrom found that in 1965-1985, respondents who said they worked 40-44 hours a week overestimated their work hours relative to the diary measurement by about 2 hours per week, and the overestimation grew as hours increased — respondents who said they worked 55-59 hours a week overestimated by 10 hours per week.

Robinson and Bostrom also found that women had larger overestimates than men, with women working more than 20 hours per week overestimating their work week by approximately 7 hours per week compared to 3 hours for men.

Robinson and Bostrom used time use surveys from 1965, 1975, and 1985. I was curious how their data would replicate using more recent ATUS data (2015-2022, except 2020).

Among all employed respondents, their CPS estimate of how much they worked last week almost exactly matched their ATUS calculated work time — less than 15 minutes over the course of the week — but there was quite a difference between work week categories. Respondents reporting that they worked between 35 and 44 hours last week had diaries that closely matched their reporting, while differences grew at higher estimated work weeks.

There was a small difference between men and women, with men slightly underreporting their work hours (-0.5) and women slightly overreporting (1.0), but again there was categorical heterogeneity.

Children and Education

William Sundstrom had a follow up paper in 1999 discussing potential causes for the overestimation as well as the potential for mismeasurement, but this caught my eye: “the presence of children under 5 was associated with much greater overestimation of work hours among women.”²

In the more recent data I downloaded, respondents with a child under 6 (regardless of sex) overestimated their work hours by about 1.1 hours, compared to no difference for respondents without a young child. Women with young children overestimated by 2.4 hours, while men with young children had no difference. The general trend of increasing overestimates with increasing estimated work hours continued, as shown in the graph below.³

Sundstrom also found that workers with higher levels of education reported larger work week overestimates.

In more recent surveys, estimates remained pretty close regardless of education, with respondents without a high school diploma underestimating work hours by 1.6 hours per week (men, -3.2; women, 0.7), and respondents with graduate degrees overestimating work hours by 0.8 hours (men, 0.6; women, 1.0).

Making Motherhood Work

This brings me to an interesting book by Dr. Caitlyn Collins, Making Motherhood Work: How Women Manage Careers and Caregiving, which I was able to pick up at the library. Dr. Collins is a professor of sociology at Washington University in St. Louis, and her book was published in 2019. I can’t remember how I first heard of her work, but it was probably in one of the many articles I like to click on that have some relationship to time use and American family life.

Like many professors, she includes a detailed methodology appendix and citations.⁴ Collins is quite explicit that her work focuses on “middle class” mothers. Many of the 32 women she interviewed in the US work in management. She notes in the appendix that “even women working as administrative assistants reported incomes over $60,000 a year,” when the median wage for administrative assistants and secretaries in 2022 was just under $45,000.

The vast majority of Collins’ interviewees also report working more than 45 hours per week. As I was reading the chapter, I wanted to know if there was a way to calculate their likely “true” hours, if their time diaries followed the national norm.

The first problem I ran into was approximating Collins’ use of “middle class” onto the ATUS samples. She notes in the appendix that she used factors like income, spouse’s income, education, occupation, and self-reported class to come to a middle class framework, but she also notes that the administrative assistants in her sample were making $60,000 per year ($72,000 adjusting from 2019 to 2024 dollars). The median personal income of her interviewees was $90,000-$99,999.

Meanwhile, median annual earnings for women with children in the ATUS sample was $44,188 in 2024 dollars, suggesting that Collins’ sample doesn’t necessarily map to the middle quartiles of US earners. Even restricting the ATUS sample to women with bachelor’s degrees or higher yields annual earnings of $71,223.

Ultimately, I restricted my comparison sample to women with young children, with a bachelor’s or graduate degree (all but 3 of Collins’ interviewees had at least a bachelor’s degree, and almost half had a graduate degree), and earning between 40th and 90th percentile wages ($61,365-$147,982, median $88,415) for this group. Collins’ median age was 38, my ATUS sample’s median was 41. 78% of Collins’ sample was married, compared to 77% of the ATUS comparison sample.

Using the ATUS sample and assuming a linear relationship, a mother similar to Collins' sample who gives a survey questionnaire estimate of 40 hours of work per week is likely working 38.3 hours on a time diary basis, while a respondent who estimates 55 hours of work per week is likely working 46.3 hours. On a questionnaire basis, a respondent who estimates she works 55 hours per week is estimating that she works 37.5% more than a typical work week, whereas on a diary basis it’s more like 20.9%.

Why the Difference?

I have two possible explanations to add or expand on from the 1994 and 1999 papers, specifically for mothers. The first is the difference between primary and secondary activities — mothers may have been more likely to report primary childcare on their time diary during work hours, especially for longer work weeks that might spill more into evenings and weekends. Because ATUS does not collect secondary market work activity, this time might be captured by survey question estimates but missed by time diary measurements.

The second is leave (paid or unpaid) — if you usually work 40 hours a week and I ask you how many hours you worked last week, but you took 2 hours off Tuesday morning to go to the dentist, do you say 40 or 38? You might say either. But a time diary would note the 2 hours off. Mothers, particularly of young children, may be more likely to take time off for sick care or doctors appointments, and by restricting the sample to higher-earning and higher-education mothers, I’m more likely to pull women with paid sick and vacation leave. A woman who gets 160 hours of paid leave per year (high, but not unheard of) and takes her whole leave would average about 37 hours of paid work per week on a diary basis, but would possibly respond that she works 40 hours per week, since that’s her usual schedule.

What About the Dads?

Using the same sample selection criteria but for changing only sex, a father who reported working 40 hours per week could be estimated to be working 42 hours by diary measurements, whereas a father who estimates he works 55 hours per week is likely working 47.9, 14% more than the diary measurement of a father working a typical full time work week.

Do High-Earners (Claim to) Work More?

A bit! First decile earners (mean earnings: $12,435 annually) average 29.7 hours of work in diary measurements and under-report their hours in survey question responses by 3.3 hours. Tenth decile earners (mean: $141,573) average 43.4 hours of work and over-report their hours by about 1.5 hours.

Overestimating Housework

Despite the previous paragraphs, Bianchi, Robinson, and Milkie spend relatively little time on paid work measurement differentials. Chapter 2 of Changing Rhythms of American Family Life instead focuses on estimate versus diary measurements of housework.

The authors found that, like paid work, estimated hours of housework were larger than diary measurements. According to Table 2.5, in the 1998-1999 data, men reported that they spent an average of 21 hours per week on housework, but according to their time diaries, spent 10.2 hours per week on housework as a primary activity or 12.1 hours per week on housework as a primary or secondary activity. Women reported an average of 34.5 hours per week on housework, but 14.7 hours on primary housework or 16.6 hours on primary and secondary housework in diary measurements.

Unfortunately, the modern ATUS does not track secondary activities except for child care and eating. It also only collects survey question estimates for paid work. The 1998-1999 data available through the American Heritage Time Use Survey Extract Builder also does not keep all the variables from the original survey, which is behind the paywall at the Inter-university Consortium for Political and Social Research.⁵

Nevertheless, here are some graphs on more current housework time use.

Jean Kimmel helpfully defines housework to “[include] all activities that could have been purchased in the marketplace, if a household member had not spent time doing them.”⁶ In 2019, the median hourly pay for maids and housekeeping cleaners was quite low — $11.95 per hour. Assuming this low hourly pay and a marginal tax rate of 22% (7.65% of payroll taxes and just under 15% for federal and state income taxes), the cost for an employed man to pay to substitute his housework with market work would be $140 per week, and for a non-employed woman it would be $417 per week ($15.32 per hour for both, about a twenty-fifth percentile hourly wage in 2019).

Relatedly, Agatha Christie used to spend 5.2% of her household income on a live-in maid.

Why Agatha Christie could afford a maid and a nanny but not a car

1. An interesting micro-example of this is parental reading to children, where survey question estimates are much higher than time diary estimates, likely due to social desirability bias: https://www.jstor.org/stable/25046699

2. Sundstrom, William A., The Overworked American or the Overestimated Workweek? Trend and Bias in Recent Estimates of Weekly Work Hours in the United States (November 1999). Available at SSRN: https://ssrn.com/abstract=1104054 or http://dx.doi.org/10.2139/ssrn.1104054

3. I removed some longer work week categories as the sample sizes shrunk

4. Journalists, especially those publishing full-length books, could learn something here. Ahem. Also, if you are going to publish a popular article drawing on a book, read the methodology. There are often important limits discussed in the methodology. Collins notes on pg. 278: “None of my interviewees were poor, and all worked in professional or semi-professional contexts.”

5. Even though it was funded with federal tax dollars. Apparently FOIA only applies if I can point to a regulation or administrative order that somehow cited the data in question.

6. Jean Kimmel, How Do We Spend Our Time?: Evidence From the American Time Use Survey (W.E. Upjohn Institute, 2008), 64.