The trends observed in Table 2 (the disproportionate decline in the work hours of low wage workers between the 1890s and 1973 and the increase in the work hours of high paid workers and the decrease in hours of low paid workers between 1973 and 1991) persist even within groups of male workers.11 Within wage deciles the lower paid workers worked the longest day in the 1890s whereas the higher paid workers worked the shortest day in 1991. Furthermore, the relationship between daily hours and the wage rate observed in Table 2 is seen within all occupation and industry categories (see Tables 4 and 5), implying that the pattern is not solely due to changes on the factory floor. The pattern persists within age groups and within occupation and industry groups controlling for age, marital status, number of dependents, and state and year fixed effects as well. Between 1973 and 1991 the dispersion in hours by wage decile among men working less than 40 hours a week (and hence not subject to legal overtime provisions) widened, suggesting that disproportionate increases in overtime rates of pay cannot explain the changing hours pattern.
For workers for whom I have only a yearly wage, I estimated the length of the work year assuming a work year of 307 days (6 holidays and Sundays off) minus the number of days lost due to ill health, unemployment, or other factors. The last two imputation procedures introduce systematic bias, but, by examining workers paid by the hour, I am able to assess the likely effect of this bias on estimates of the wage elasticity of daily hours worked. Workers for whom the only wage information is the amount paid by the ton, mile, or piece were deleted from the sample. All wages were adjusted to be in real 1895 dollars. Using my second wage variable the mean wage in the sample was 19 cents per hour and for workers paid by the hour it was 23 cents per hour. The sample mean is therefore close to the national mean hourly wage for manufacturing workers of 20 cents per hour in 1895 (Series D 765-778 in U.S. Bureau of the Census 1975: 168).
Questions on hours of work comparable to those in the pooled datasets were asked in a supplement to the 1991 Current Population Survey. Mean hours of work per day (5 days a week) were 8.6 for men and 7.7 for women. Fifty-seven percent of the men in the sample stated that they worked eight hours a day (see Figure 1). The most common pattern was for work to begin at 8 am and end at 5 pm. A comparison of work start and end times and the reported length of the work day suggests that the majority of workers excluded lunch breaks from reported daily hours of work. The questions asked in the 1973 Current Population Survey were somewhat different and usual hours per day were estimated from usual hours per week divided by usual days per week. When similar information is used to estimate hours per day in the 1991 data, the average length of the usual working day falls slightly to 8.4 hours for men and 7.6 for women. The reported work day may include overtime if overtime was “usual” but because the proportion of low wage to high wage decile workers receiving overtime pay did not change between 1973 and 1991, changes in overtime coverage are unlikely to bias my results. Although time diary studies suggest that the length of the work day is overestimated both in 1991 and in 1973, but particularly in 1991, this will not bias my estimates of changes in relative hours of work.
Despite a slight increase in the coefficient of variation of daily hours worked from the 1890s to the present (see Table 1), for men the distribution between the 90th and 10th percentiles has become more compressed because the majority now work an eight hour day (see Figure 1). However, for women the distribution first narrowed between the 1890s and 1973 and then widened between 1973 and 1991, largely because the widening of the distribution for full-time workers outweighted the narrowing of the distribution for part-time workers.
The final dataset contains over 11,000 men aged 25 to 64 and over 1,100 women aged 18 to 64. Although the men in the survey are predominately upper working class non-farm wage earners and the women manufacturing operatives, there is enough variation in the data to reweight by broad occupation or industry category. As expected, less than 10 percent of women were married. Although unionized workers are oversampled, unionization lowered hours of work by only 2 percent, suggesting that this will not bias my results.
The questions that were asked about hours of work varied slightly by state, but all referred to usual hours of work per day. I assume throughout that the usual work day excludes lunch time, breaks, and overtime, but includes time spent on the job not working. None of the states had hours legislation at the dates of the surveys.
This paper uses micro-level data to examine the distribution of daily hours of work in the 1890s and uses comparable data from 1973 and 1991 to examine how inequality in the length of the work day by the hourly wage has changed. Although only data on the hours of men are strictly comparable over time (because of increases women’s labor force participation throughout this period), results are presented for women and married couples as well. The paper first describes the data, then presents tabulations of the length of the work day by the hourly wage in the 1890s, 1973, and 1991, and discusses the factors that might have affected the distribution of hours, such as hours legislation and decreases in the number of daily hours that workers are willing to supply and that firms demand from each worker. The paper will show that in the past the labor supply curve was much more backwards bending and will examine why. The paper concludes with an analysis of the implications of the findings for earnings inequality.
The length of the work day fell sharply between the 1880s when the typical worker labored ten hours a day six days a week and 1920 when his counterpart worked an eight hour day six days a week. By 1940 the typical work schedule was eight hours a day five days a week. Although further reductions in work time largely took the form of increases in vacations, holidays, sick days, personal leave, and earlier retirement, time diary studies suggest that the work day has continued to trend downwards to less than eight hours a day. This decline in work hours, unmeasured by such common indicators of well-being as income per capita, surely represents one of the larger increases in the standard of living during this century.
Examining 911 services also provides a glimpse into the challenges (and types of data) which are necessary for accurate measurement of productivity in the service sector. In particular, service sector productivity measurement must incorporate the quality of the activity (such as timeliness) as well as whether the services received by the customer are responsive to his idiosyncratic characteristics (in this case, different patients experience different diagnoses and different degrees of severity of illness). By developing and analyzing a novel dataset, we are able to provide evidence about both of these factors (in this case, timely response and allocation of patients to appropriate hospitals). Of course, we are not the first to evaluate multiple attributes of a service provided. However, our analysis is further able to connect these measures of quality to a well-defined overall service outcome measure, mortality.
From our analysis in this paper, we draw several conclusions which we hope will impact future research. First, our results highlight that emergency response systems play two distinct roles: productive and allocative. It therefore seems important to consider the potential bias which arises in studies which take allocation as exogenous or which do not account for the heterogeneity in county mortality rates which are induced by higher levels of pre-hospital care (such as lower response times or on-the-scene defibrillation). Further, the incentives generated by the pre-hospital system need to be taken into account when regulators and insurance companies consider creating additional incentives for hospitals. Our analysis highlights one particularly important feature of the pre-hospital system: it interacts with the incentives of hospitals to adopt new technologies and maintain highly rated emergency facilities.
Our final empirical exercise considers directly the incentives of hospitals to adopt higher levels of technology. Identifying the role that hospital characteristics play in determining the allocation of ambulance patients is in many ways similar to a study of a differentiated goods demand system, in which hospitals compete in the marketplace for patients on the basis of geography and characteristics. However, these two settings also differ in some respects; in particular, while hospitals will presumably have incentives to attract some ambulance patients, a given hospital may want to deter particular types of patients (the uninsured or patients whom are hard-to-treat but do not generate significant income). While these distributional questions are extremely interesting, the present analysis will focus on the sensitivity of the overall patient share to particular hospital investments.
We further find that, excluding the largest counties, patients with very severe and very mild indications were most likely to go to hospitals with high levels of technology. The result for less severe patients could be due to the use of ambulances for cases which are more elective in nature, since patients may be reporting emergencies in order to have access to the ambulances for basic transportation. Patient insurance status further affects the hospital allocation decision. We find that privately insured patients are allocated in a similar fashion to Medicare patients. However, Medicaid and self-pay patients are more likely to be treated in high-tech hospitals.