The presentation presents a detailed review of a paper published by Peter Kuhn. We start by examining the overall labour trend during this period of 1979 to 2006. We will then methodically try to deconstruct the trend to attempt to reveal the true mechanism that is causing this phenomenon.
1. THE EXPANDING
WORKWEEK
UNDERSTANDING TRENDS IN LONG WORK HOURS
AMONG US MEN, 1979-2006
PAPER BY: PETER KHUN & FERNANDO LOZANO
PRESENTED BY: JAY LIN
2. INTRODUCTION
• Between 1970-1990, US men who worked 48+ hours
per week rose from 15.4% to 23.3%. (Census
reference week)
• Between the 1980 Census and 2005 ACS, the share
working 48+ hours rose from 16.6% to 24.3%.
• Recent increase is of interest for understanding
long-term trends in work behavior.
3. DATA SOURCES
• Outgoing Rotation Groups of the Current
Population Survey from 1979 to 2006.
• Better wage information
• Consistent hours measures across many years
• Large sample size
• Information on method of pay (Salary vs. Hourly)
• 7 decades of US Census microdata
• 2005 American Community Survey
4. SCOPE
• Focus on men aged 25-64 but not self-employed
• Focus on weekly hours instead of annual hours
• Filters out the effect of technical innovations
• Subtracts vacations and other forms of leave
• Focus on a particular feature of weekly hours
distribution
• Fraction of 30+ usual hours workers usually working 50+ hours
or more
9. CPS REDESIGN
• CPS underwent major redesign in 1994
• Could this redesign explain the results the long-term
trends?
• No. Why?
• Main job-hours question did not change
• Majority of increase in long hours occurs before 1994
• Similar pattern in men’s long hours in General Social Survey
10. FRACTION OF MEN WORKING 50+ HRS
Table 1A 1979 1989 2000 2006
All Men 0.161 0.193 0.190 0.178
Full Time Men 0.164 0.199 0.207 0.195
Among FT Men:
Salaried 0.244 0.312 0.320 0.301
Hourly 0.086 0.094 0.105 0.096
11. FRACTION OF MEN WORKING 50+ HRS
Table 1B 1979 1989 2000 2006
Among FT Men:
Age 25-34 0.171 0.197 0.196 0.167
Age 35-44 0.185 0.221 0.222 0.208
Age 45-54 0.154 0.193 0.216 0.213
Age 55-64 0.128 0.154 0.178 0.191
12. FRACTION OF MEN WORKING 50+ HRS
Table 1C 1979 1989 2000 2006
Among FT Men:
Less than HS 0.124 0.121 0.116 0.099
HS Graduate 0.137 0.155 0.149 0.153
Some College 0.166 0.190 0.194 0.182
College Grad. 0.240 0.303 0.312 0.278
13. FRACTION OF MEN WORKING 50+ HRS
Table 1D 1979 1989 2000 2006
Among FT Men:
1 (highest wage) 0.151 0.243 0.297 0.268
2 0.137 0.193 0.214 0.219
3 0.132 0.176 0.199 0.189
4 0.176 0.202 0.184 0.172
5 (lowest wage) 0.217 0.186 0.151 0.133
16. HOW DO WE JUSTIFY THIS
TREND?
NEED TO LOOK FROM A DIFFERENT ANGLE
17. TYPES OF LABOUR SUPPLY VARIATION
• Moonlighting
• Increases in long work hours were offset by fewer hours
worked in other job?
• Employment Rates
• Concentration hypothesis
• Concentration of intense work activity in the 1980’s
• Separated by more bouts of inactivity, low activity, and/or
earlier retirement
18. MEN’S LABOR SUPPLY BY EDUC
Table 2A 1979 1989 2000 2006
Ages 25-64:
% of Men Employed
Less than HS 0.763 0.709 0.724 0.726
High School Grad. 0.892 0.859 0.831 0.807
Some College 0.904 0.892 0.871 0.843
College Graduate 0.940 0.928 0.914 0.890
19. MEN’S LABOR SUPPLY BY EDUC
Table 2B 1979 1989 2000 2006
Ages 45-54:
% of Men Employed
Less than HS 0.814 0.769 0.700 0.710
High School Grad. 0.910 0.886 0.837 0.825
Some College 0.920 0.911 0.874 0.863
College Graduate 0.961 0.943 0.935 0.929
20. MEN’S LABOR SUPPLY BY EDUC
Table 2C 1979 1989 2000 2006
% of Men Employed Working Long Hours
Less than HS 0.111 0.126 0.111 0.110
High School Grad. 0.133 0.139 0.138 0.166
Some College 0.159 0.200 0.194 0.195
College Graduate 0.248 0.301 0.324 0.302
21. TYPES OF LABOUR SUPPLY VARIATION
• Moonlighting
• Increases in long work hours were offset by fewer hours
worked in other job?
• Employment Rates
• Concentration hypothesis
• Concentration of intense work activity in the 1980’s
• Separated by more bouts of inactivity, low activity, and/or
earlier retirement
22. TYPES OF LABOUR SUPPLY VARIATION
• Increase in Part-time work incidence
• The Census and ACS data does not support this statement
• Relatively little change over our period
• Small increase that did occur was greater among less-skilled
men
• Longer annual vacations
• BLS’s Employee Benefits Survey show very little trend
between 1980 and 1997
• Includes all dimensions of paid leisure
23. WHAT DOES THIS MEAN?
• Concentration hypothesis does not explain this
trend
• Other dimensions of labour supply variation also do
not support this hypothesis
24. WHAT WE HAVE SEEN SO FAR…
• Increase in long hours was strongest before 1990
• Concentrated among highly educated, high-wage,
and older men
• Confined to workers paid on a salaried basis
• What else can we use to try and decipher this
trend?
26. DEMOGRAPHIC SHIFTS
• Optimal work hours vary across types of workers
and types of jobs
• Long-term changes in mix of workers and jobs in
labour force could explain increase in long hours
• Assess the importance of such factors in
determining our trends in long work hours
27. DEMOGRAPHIC SHIFTS
• Restrict our attention to the period from 1983 to
2002
• Pool our observations to generate adequate
sample sizes.
• 1983-1985 will represent the beginning of this period
• 2000-2002 will represent the end of this period
• This period contains all of the long term increase in
long hours observed in our data
28. LINEAR PROBABILITY MODEL
Table 3 Regression Coefficients Sample Means
1983-1985 2000-2002 1983-1985 2000-2002
HS Graduate .006* (.002) .007* (.003) .344 .310
Some College .014** (.003) .025** (.003) .215 .265
College Grad. .040** (.003) .065** (.004) .268 .314
Age 35-44 .001 (.002) .008** (.002) .285 .332
Age 45-54 -.020** (.002) -.007** (.002) .195 .261
Age 55+ -.039** (.002) -.030** (.003) .131 .115
Salaried .113** (.002) .130** (.002) .501 .483
Married .018** (.002) .027** (.002) .761 .672
Union -.029** (.002) .016** (.002) .275 .172
Observations 213,062 210,640
R-squared .134 .127
29. FRACTION OF MEN WORKING 50+ HRS
Table 1C 1979 1989 2000 2006
Among FT Men:
Less than HS 0.124 0.121 0.116 0.099
HS Graduate 0.137 0.155 0.149 0.153
Some College 0.166 0.190 0.194 0.182
College Grad. 0.240 0.303 0.312 0.278
30. LINEAR PROBABILITY MODEL
Table 3 Regression Coefficients Sample Means
1983-1985 2000-2002 1983-1985 2000-2002
HS Graduate .006* (.002) .007* (.003) .344 .310
Some College .014** (.003) .025** (.003) .215 .265
College Grad. .040** (.003) .065** (.004) .268 .314
Age 35-44 .001 (.002) .008** (.002) .285 .332
Age 45-54 -.020** (.002) -.007** (.002) .195 .261
Age 55+ -.039** (.002) -.030** (.003) .131 .115
Salaried .113** (.002) .130** (.002) .501 .483
Married .018** (.002) .027** (.002) .761 .672
Union -.029** (.002) .016** (.002) .275 .172
Observations 213,062 210,640
R-squared .134 .127
31. FRACTION OF MEN WORKING 50+ HRS
Table 1B 1979 1989 2000 2006
Among FT Men:
Age 25-34 0.171 0.197 0.196 0.167
Age 35-44 0.185 0.221 0.222 0.208
Age 45-54 0.154 0.193 0.216 0.213
Age 55-64 0.128 0.154 0.178 0.191
32. LINEAR PROBABILITY MODEL
Table 3 Regression Coefficients Sample Means
1983-1985 2000-2002 1983-1985 2000-2002
HS Graduate .006* (.002) .007* (.003) .344 .310
Some College .014** (.003) .025** (.003) .215 .265
College Grad. .040** (.003) .065** (.004) .268 .314
Age 35-44 .001 (.002) .008** (.002) .285 .332
Age 45-54 -.020** (.002) -.007** (.002) .195 .261
Age 55+ -.039** (.002) -.030** (.003) .131 .115
Salaried .113** (.002) .130** (.002) .501 .483
Married .018** (.002) .027** (.002) .761 .672
Union -.029** (.002) .016** (.002) .275 .172
Observations 213,062 210,640
R-squared .134 .127
33. PREDICTED SHARE OF MEN WORKING
LONG HOURS
Table 4 83/85 00/02 83/85 00/02
Coefficients Coefficients Coefficients Coefficients
& & & &
83/85 Means 00/02 Means 00/02 Means 83/85 Means
(1) (2) (3) (4)
All FT Men .166 .196 .173 .194
Salaried FT Men .253 .298 .271 .282
• According to this Oaxaca analysis
• Using 83/85 regressions:
• 0.7% of 3.0% explained for all FT Men
• 1.8% of 3.0% explained for salaried FT Men
• Using 00/02 regressions:
• 0.2% of 4.5% explained for all FT Men
• 1.6% of 4.5% explained for salaried FT Men
34. THIS TELLS US THAT…
• Although these observed characteristics played a
role to explain the increase in long work hours
• But the majority of the recent increase cannot be
account for by these factors
• Are there problems with this analysis?
35. DETAILED INDUSTRY AND
OCCUPATION MIX
• Could be possible that our previous analysis failed
to capture the true effect due to the broad
categories?
• We can conduct a more detailed Oaxaca
decomposition to see if that is the case
36. PREDICTED SHARE OF MEN WORKING
LONG HOURS – DETAILED MIX
Table 5 83/85 Cell Means
&
00/02 Cell Means
&
83/85 Cell Means
&
00/02 Cell Means
&
83/85 Mix 00/02 Mix 00/02 Mix 83/85 Mix
(1) (2) (3) (4)
All FT Men
By Occupation .163 .204 .170 .198
By Industry .163 .203 .170 .200
Salaried FT Men
By Occupation .252 .320 .258 .315
By Industry .257 .325 .262 .318
Note: Restricted attention to cells with 50 or more observations in both
sample periods. Reduces down to 315 occupations and 201 industries.
37. COMPOSITION EFFECTS
• Table 5 strongly indicates that long hours increase
occurred within very detailed occupation and
industry groups
• Table 5 also reinforces that the variation in the mix
of jobs performed cannot fully account for this
increase in hours
39. CHANGES IN WORK INCENTIVES
• We have already considered:
• Measurement techniques
• Composition effects
• Want to look at effects of changes in financial
incentives to explain the increase
• Consider 3 proxies that can be calculated from the
CPS data
40. THREE PROXY VARIABLES
• Level of real average earnings
• JMT model
• Long hours premium
• Cross-sectional relationship between usual weekly hours
and total weekly earnings.
• Within-group earnings dispersion
• Bell and Freeman argue that if there are more “prizes”
available to workers then it should elicit more work hours
• Dangling the carrot and holding back the stick
41. AVERAGE HOURLY EARNINGS
• Could use the model proposed by (JMT, 1991) and
(Juhn, 1992)
• States that +ve uncompensated labour supply
response to real wages can be explained by men’s
employment rates
43. AVERAGE HOURLY EARNINGS
• This model faces difficulties in explaining the trends
of long hours in the 1980’s and 1990’s
• Could variants of the JMT model provide more
clarity?
• Invoke CPI bias as a solution to our result?
44. LONG HOURS PREMIUM
• Focus on the trends in the marginal incentives to
supply hours beyond the standard full time amount
• Straightforward for hourly workers but how would
we account for the bulk of the rise in hours for
salaried individuals?
• Intuitively, it would be a poor measure for salaried
workers.
45. GAINS FOR SALARIED MEN
• Types of financial rewards:
• Earning a bonus or a raise
• Earning a promotion
• Signaling to labor marketing that you are
productive/ambitious
• Acquiring extra skills/networks/contacts
• Enhanced prospect of keeping one’s current job
• Measuring the above rewards is a challenge and
especially with cross-sectional data sets like CPS
46. HOW DO WE MEASURE IT?
• Cross-sectional relationship, within labour market
subgroup, between usual weekly hours and total
weekly earnings
• Assume that tastes for work are relatively fixed
personal characteristics
• Intuitively, we can see that hourly workers would be
more strongly associated with this measure then
salaried workers
50. WITHIN-GROUP EARNINGS DISPERSION
• Increase in within-occupation earnings dispersion
means that there are variety of rates of pay
• Move from standard rates of pay to potentially
performance-related pay
• Can include bonuses, stock options, job security,
size of pay between firms
52. WITHIN-GROUP EARNINGS DISPERSION
• Previous table shows a substantial increase in this
proxy for marginal work incentives
• Increase is substantially greater among salaried
men than hourly paid workers
• Good candidate to explain the increase in men’s
long work hours
53. DISAGGREGATED ANALYSIS
• Looking at the changes in long work hours among
salaried men
• Conduct a disaggregated analysis of long-term
(roughly decadal) trends in long hours
• Want to study the association between changes in
measured work incentives across subgroups of full-
time salaried men.
55. DISAGGREGATED ANALYSIS
2 Digit 3 Digit 2 Digit 3 Digit
Industry Industry Occup. Occup.
D. Changes in Earnings Dispersion
Change in Standard Deviation of 0.143 0.145** 0.128* 0.096**
Log Earnings (0.088) (0.047) (0.068) (0.039)
Change in the Standard Deviation 0.160* 0.151** 0.143** 0.105**
of Log Wage (0.088) (0.048) (0.070) (0.039)
Change in the Standard Deviation 0.082 0.087* 0.170** 0.105***
of Salary Residual (0.110) (0.049) (0.073) (0.039)
Change in the 90-10 Salary 0.071 0.083** 0.098** 0.064**
Residual Gap (0.062) (0.030) (0.040) (0.023
Observations 90 369 86 416
56. DISAGGREGATED ANALYSIS
• Provides some evidence that long hours incidence
over this period were more common in occupations
and industries where increases in average hourly
earnings were lowest and where within-group pay
inequality increased the most
• Suggests a negative income effect of higher pay on
labor supply
• Individualized pay provides greater marginal
incentives for salaried workers to supply extra hours
57. CONCLUSION
• We showed that employed American men are
more like to work 50+ hours per week today than a
quarter century ago
• We ruled our a number of possible causes such as
CPS survey techniques, cyclical phenomenon, and
mix of occupations/industries
• Also cannot be interpreted as simple reallocation of
a fixed amount of labor
We want to try and investigate why employed American men are more likely to put in long work weeks today than a quarter century ago?We will want to identify the time periods and the parts of the male labor force where the recent increase was the strongestWe will want to also try to come up with an explanation as to why this is the case
These are cross sectional data sets that we are pooling over the time period.The American Community Survey (ACS) is an ongoing survey that provides data every year and collects information on key indicators such has your personal information and employment informationCensus is taken every 10 yearsSo for the context of this paper, we have to use both sets of data to help us cover the time period that we want to evaluate
1.This is usually the prime working age for individuals in the western world. This is where the recent increase was the strongest.2.Technical innovations have an impact on internet job matching services and growth of the temporary help industry. They increase the efficiency of labor market matching.They are of no interest to us in this paper so we use a measure of weekly hours to filter out these effects. Also our real interest here is in the labor supply choices made voluntarily by workers.Also annual work weeks information suffers from a serious measurement problem in that it doesn’t subtract vacations and other forma of leave from measured work time.3.
Using data going back to the 1940’s, we can see our story beginning to developJust under 25% of men in our data set worked 49+ hours in 1940Dipped to it’s lowest point in 1970Our area of interest is starting in 1980 where the trend sharply increases from 20% to just over 25% in the short span of 20 yearsNow we could ask ourselves, entrepreneurs in our sample are generally know for working more hours and maybe that subgroup is showing carrying this trend that we are seeing?
Subtracting this subgroup from our dataset, we can still see that the trend we saw is still prevalentSo even with the entrepreneur effect filtered out, we can still see this trend of men working longer hours between the 1980’s to 2000The effect is even larger then in the previous graphHow can we try to explain this phenomenon?Before we go into that, we should take a look at a few more graphs to help us build a better picture
Looking at just our specific period of interest, we can see in more detail the % of males that were working long hours. Really peaked at 1995 and held until after 2000 when it started to drop off
It would also be of interest to look at the employment population ratio over this period of time as well to see if there is a connection between themCan see that the long term employment population ratio had a downward trend.Two interesting things to note is that long hours worked decreased during the recessionary periods of 1983, 1992, and 2002.1983: combination of 1979 energy crisis, stagflation, and contractionary monetary policy from the US federal reserve1992: combination of the 1990 spike in oil prices, 1987 stock market collapse (black Monday), savings and loan crisis
So that rules out that survey techniques played a role in the our data.Similar trends are also seen using other nationally representative data sets general social survey with a larger increase in the 80’s then the 90’sAnother possible issue that we have to rule out is that this could just be caused by business cycle effects so let’s take a look at that
The following tables, we will look in more detail on the size and distribution of the increase in men’s long work hoursWe pick the similar points in the business cycle to conduct our analysis, during the peak employment rates right before the recessionary periods began.All Men: increased modestlyFT Men (30+ hrs): show the same trend with a bigger increase among this subgroup then the general populationSplitting FT men into two further groupsSalaried: 5.7% increaseHourly: 1% increaseAlso note that the magnitude of proportion of those that work between these two different classes.Salaried men at all time periods had a higher fraction that worked 50+ hours
Breaking it down by age groups, we can see that the youngest age group had a decreasing trendWhile every other age group experienced an upward trendThe biggest jump was for the 45-54 age group which increased by 6.3%
Looking from another angle, we can see what the spread is like across different educational backgroundsLess than HS individuals had -2.3%HS graduates had 2.0% increaseSome college had 2.4% increaseCollege grad had 3.8%Again we are see this similar trend where the most educated tend to work longer on avg then those less educated
Looking at this trend from wage quintilesAgain we can see that the highest wage earners also had the most substantial increase in long hours (11.7%)While the lowest wage earners had a substantial decrease (8.4%)Incidence of long hours increased as the wage rises is what we are seeing hereGraphically…
Graphically, we can see this result and now along with the following graph…
This graph shows the wage quintiles plotted against their avg wage across timeCan see that the only meaningful wage growth occurred in the highest quintile/wage earnersEveryone else experienced negative wage growthTwo things to note from these two graphsIncrease in long hours was strongest among the quintile to experience real wage growth in this periodThat this increase was strong during the period when there was real wage growthLast thing to note, most striking feature is that in 1979 bottom 20 was willing to work more long hours then the top 20But by 2006 this situation had completely reversed. This feature forms part of the recent change in men’s labor supply behavior that we attempt to understand in this paper
Remember that we saw that employment to population ratio fell over the period that we are looking atThen this raises a possibility that the increase in long work week is just offset by decreases in other dimensions of labor supplySo that people are working more during the week due to other factors other factors that has nothing to do with their main job
MoonlightingBecause the main indicator of works hours used refers only to the respondent’s “main” job then could it mean that for those that held more then one job that they would just reduce their work hours for those jobs?To address this issue, we examined the May CPS surveys in 1979, 1991, and 2001 because they contained information on multiple job holdings and usual hours worked in all jobs.You can believe me when I say that this analysis shows a similar increase in works hours when all jobs are taken into account Analysis showed that there was little change in either the rate of multiple job holding or the incidence of long work hours among multiple jobholdersSo then moonlighting won’t be able to explain the increase in long hoursThe analysis showed that the entire increase in long total work hours was in the usual hours of workers who held only one jobEmployment RatesConcentration hypothesis: individuals will work hours will fluctuate over time with bouts of intense work separated by bouts of low to no activityConsidering our result from when we addressed moonlighting, it could be argued that the computer programming is most efficient when done in long sessions or that managerial jobs are increasingly very intense followed by early retirement
We can try to see if there is evidence for thislabour supply variation look at men’s labour supply by educationThis first part of the table tells us the share of men employed according to their highest level of complete educationLooking at all age groups, men’s labour supply dropped across all age groups with the biggest drops in the two middle categoriesTo test this hypothesis, we can look at 1979-1989 where the rise in long hours occurred mostly, we would think that the share of men employed would be relatively the same during this period while people worked more long hours and stuff.
We can see pattern more dramatically if we look at just the 45-54 age groupEmployment rates fell dramatically over this period for this group and again focusing on the 1979-89 period, employment rates go down while long hours increased.
Examining the data that we used to build table 1 that we saw earlier, we saw that there was relatively little change over our period in the share of men working part timeBLS = Bureau of Labor StatisticsIncludes all dimensions of paid leisure: annual vacation days, holidays, paid lunch, rest time, and paid sick leave
In sum, we can conclude that variations in the dimensions of labor supply do not support the concentration hypothesis and explain the increase in long hours in our time period.
Just to review what we have learned so far from the different descriptive statistics that we have seen so farSo the case that we are seeing is that while overall employment fell this period, it fell the least among the most skilled group of workers and they experienced the most increase of long hours worked.We aren’t too concerned about hourly workers because generally hours worked is highly correlated with wage.
Now we can try to explain this trend through the different makeup and characteristics of our population over time or the mix and characteristics of jobs and industry.Could these changes over time cause the increase in long hours?
Marginal productivity and marginal disutility of an extra work hour are likely to vary as well among different workersWe will attempt to decompose these trends in long work hours to assess the important of such factors
We restrict out attention to this period because the three digit occupation and industry codes changed dramatically both in 1983 and near the end of our sample periodWe will only evaluate cells that have 50 or more observations combined for all three time periods at each pointRecall that this period contained almost all of our long term increase in long hours observed in our dataWe will use a linear probability model for the incidence of working long hours regression
Looking from another angle, we can see what the spread is like across different educational backgroundsLess than HS individuals had -2.3%HS graduates had 2.0% increaseSome college had 2.4% increaseCollege grad had 3.8%Again we are see this similar trend where the most educated tend to work longer on avg then those less educated
We can see that older men were more likely to work less in both periodsThis pure age effect was hidden in table 1 because older men are more likely to be salaried and married which are characteristics that contribute to higher hoursConsistent with table 1 however is the negative impact of age on propensity to work long hours does weaken slightly between periods
Breaking it down by age groups, we can see that the youngest age group had a decreasing trendWhile every other age group experienced an upward trendThe biggest jump was for the 45-54 age group which increased by 6.3%
So table 3 shows that the population of working american men became better educated, less married, and much less unionized during our sample period.
In this table we see the results of the Oaxaca decomposition of changes in long work hours
To see how table 9 works, consider row 1It shows that 2 digit industries with higher average hourly earnings at the beginning of a period experienced greater growth in long work hours during that period.The same is true for the other disaggreations: 3 digit industries, 2 and 3 digit occupationsAccording to row 2, the same is true when we substitute the beginning of period level of total weekly earning for the level of average hourly earnings.Together rows 1 and 2 of this table confirm our earlier findings that the recent increase in long work hours was concentrated among high paid menRow 3 of this talbe asks whether the industries or occupations that experienced the largest increases in real hourly wage rates experienced the largest increases in long work hours.This is clear not the case; in fact the 3 digit occupation analysis shows a strong negative effectRow 4 of the table repeats this analysis using total log weekly earnings changes instead of hourly wage changesRows
The last four rows of his table examine the hypothesis that increases in within-in group residual earnings inequality were associated with increases in the incidence of long work hoursWe can see that almost all the coefficients are statistically significant
In summary, this table provides some evidence that increases in the incidence of long work horus over this period were more common in occupations and industries where increases in avg hourly earnings were lowest and where within group pay inequality increased the most.