June 2005
Vol. 6, no. 6

When employment increases, a corres-ponding rise in hours worked can usually be expected. And indeed, from 1976 to 2000, changes in employment were fairly accurately reflected in hours worked. However, since 2000, this relationship has greatly diminished (Chart A). From 2000 to 2004, employment rose 8.1%, compared with only 4.3% for overall hours worked. Such a differential is unprecedented. The robust employment growth was surprising. Lower growth in hours seemed more consistent with reduced economic growth in Canada and the stagnation of employment in the United States. This contrast led some economists to question Canada's exceptional employment record in recent years (RBC 2004).

If employment is growing more rapidly than hours, then average hours per worker are declining. According to the Labour Force Survey (LFS), average weekly hours actually worked declined for three consecutive years—2001, 2002 and 2003—before rising slightly in 2004. The decline from 2000 to 2003 affected all provinces and population groups. According to the LFS, the decrease averaged 1.4 hours per week per worker (Table 1). In annual terms, this represents a drop of some two weeks of work.

Many factors can influence hours worked. Some are structural or cyclical, such as population aging, shifts in industrial structure, the business cycle, natural disasters, legislative changes, or simply personal preference. Others originate from the survey's conceptual framework, which should be re-examined periodically to see that it is still measuring what it is supposed to. This article looks at the contribution of these different factors in the decline in hours worked.

Decomposition of actual hours

The LFS collects information on both usual hours and actual hours of work. The drop in average hours appears in actual hours.

Usual hours of work are generally more stable since they reflect regular work schedules (Chart B). Changes in usual hours reflect fairly permanent changes in weekly work schedules.

On the other hand, hours actually worked can vary from week to week. By definition, actual hours are the sum of usual and overtime hours (paid or otherwise) minus hours of absence for any reason (for example, illness, vacation, personal or family responsibilities). (For a definition of these hours, see Definitions and reference week). This relationship is expressed in the following identity:

This identity can be verified for employees only, since total hours lost are reported only for this category of workers.⁴ An examination of the average values⁵ of these components reveals that the drop in actual hours per employee happened gradually. From a peak in 2000, the rate dropped to a low in 2003 and then rallied somewhat in 2004. The 1.4 hour drop in average weekly hours between 2000 and 2003 is largely attributable to an increase in hours lost (Table 1, before adjustment). In fact, this accounted for more than 82% of the decrease in average hours.

The rest of the decrease came from a decline in usual hours, where the average fell from 35.7 to 35.5 hours per week between 2001 and 2002 and then remained stable for the rest of the observation period. Overtime hours during the period fluctuated between an average of 1.7 and 1.8 hours per week, slightly tempering the decrease in actual hours worked.

Increase in hours lost for statutory holidays suspect

A breakdown serves to identify the source of the increase in hours lost. Absences were grouped according to six types: illness, vacation, statutory holidays, personal or family responsibilities, maternity leave, and other.⁶

Between 2000 and 2003, increases occurred for all types of absence, but particularly for statutory holidays (Table 2, unadjusted data). The average number of weekly hours not worked for this reason quadrupled from 0.2 hours in 2000 to 0.8 hours in 2003. On an annual basis, this amounts to 1.4 days in 2000 compared with 5.5 days in 2003.⁷ Part of the rise occurred between 2000 and 2001, and again between 2002 and 2003. Between 2000 and 2003, statutory holidays explained almost 57% of the rise in total hours lost. However, the increase seems suspect, since the number of statutory holidays in each year was identical.

The decrease in hours: very real, but overestimated

Actual and usual hours of work are counted by the LFS every month, during a reference week that usually includes the 15^th of the month.⁸ Not all reference weeks are comparable. Among other things, some weeks include statutory holidays while others do not. These holidays have a greater or lesser effect, depending on whether they affect all Canadian workers or only some of them (for example, the Saint-Jean-Baptiste holiday in Quebec). Each year, the LFS captures only a portion of statutory holidays, and from one year to the next it captures different ones.

For example, in 2000, two statutory holidays often captured by the LFS were entirely or partially missed. Easter occurred during the week preceding the reference week and thus was not captured. Remembrance Day, which occurs on November 11 is a fixed holiday.⁹ When it falls on a Saturday, as in 2000, it has almost no effect because it will be taken on the Monday following the reference week.¹⁰ Also, the hours lost during the October 2000 reference week, which included Thanksgiving, were underestimated because of a technical problem related to the introduction of the new computer-assisted interview system.¹¹ In 2003, all these holidays were captured. A comparison between 2000 and 2003 of average actual hours per month shows the effect of the holidays not captured in 2000: hours for April, October and November 2000 appear much higher than in 2003 (Chart C). But in reality, their high value largely reflects the holidays missed by the survey. Hours in 2000 were thus overestimated because holidays were under-represented. This had a major impact on the reduction of actual hours between 2000 and 2003, since it more than doubled the decline in average actual hours.

However, even after adjusting to neutralize the estimation bias caused by the irregular presence of statutory holidays,¹² a decrease in average hours between 2000 and 2003 remains (Table 1). But it amounts to 0.7 hours on average per week per employee (just under one week annually) instead of 1.4 hours per week (two weeks annually). This decrease occurred gradually—from 34.0 in 2000 to 33.6 in 2001 to 33.5 in 2002 and then to 33.2 in 2003. What explains this decline?

Reasons for the decline

Using the adjusted data, the components of the identity were re-examined for the years 2000 and 2003. This exercise confirmed the importance of various reasons (other than statutory holidays, which are excluded from the adjusted data) in the decrease in average actual hours.

Increase in hours lost a major factor
Hours lost explained two-thirds of the drop in hours (Table 1, after adjustment) instead of the 82% before adjustment. Usual hours explained the remaining third. On the other hand, overtime, which remained steady at 1.7 hours per week, slowed the decrease in hours.

Since the distribution of employees between full- and part-time has a major impact on hours worked, the identity was done separately for full- and part-time workers (Table 3). The decrease in adjusted average actual hours between 2000 and 2003 was much greater for full-time workers—0.7 hours on average per week after adjustment, compared with a marginal 0.1 for part-time workers.

A decomposition of the drop into the components of the identity for full-time and part-time employees confirms the importance of hours lost in the drop in average actual hours (Table 3, after adjustment), accounting for 85% and 153% respectively.

Employees seeking a better balance between work and personal life
When statutory holidays are excluded, the influence of other reasons for absence between 2000 and 2003¹³ can be seen (Table 2, after adjustment). Among full-time employees, maternity leave accounted for the largest share of the increase in hours of absence—nearly one-third. This coincided with changes to Employment Insurance which, as of December 31, 2000, increased maternity, parental or adoption benefits from 30 to 50 weeks. (For further details, see Marshall 2003.) Hours of absence per female employee because of maternity leave rose from 0.6 hours per week¹⁴ in 2000 to 0.9 hours in 2003.

Again in the case of full-time employees, the second-ranking factor in the increase in hours of absence was vacation leave, which accounted for 23% of the increase. This is probably partly related to an aging workforce, as older workers are generally entitled to more leave; in fact, 70% of the increase in vacation leave was attributable to workers aged 45 and over.

By the same token, with inflation remaining at relatively low levels over the past several years, union demands have focused less on wages than on job protection (Fortin 2003) and on improvements to fringe benefits. Indeed, some employers use both wages and employment conditions to attract the best workers (Akyeampong 2002).¹⁵ The increase in vacation leave is probably also partly attributable to these new union demands, which are oriented more toward a better balance between work and personal life. The increase in leave for personal or family responsibilities—which, while more modest, nevertheless explained just over one-tenth of the overall increase in hours lost—is likely also part of this trend.

The increase in absences for 'other' reasons explained 17% of the increase in hours lost for full-time employees and reflects numerous disruptive events that occurred in 2003. Ontario was especially hard hit by the power blackout in August as well as the appearance of a number of SARS cases. For its part, British Columbia found itself in the grip of numerous forest fires and several floods, while Canadian businesses were forced to adjust to strong appreciation in the Canadian dollar.

The increase in time lost because of illness was also substantial, explaining 19% of the total increase in hours lost between the two years. Aging did not appear to be the only cause of the rise, at least for men, since only 36% of the increase in time lost because of illness was attributable to male workers aged 45 and over. For women, the figure was 70%.

For part-time employees, the increase in absences because of illness and for other reasons accounted for the lion's share of the increase in hours lost (72%). Not surprisingly, maternity leave, personal or family responsibilities, and vacations explained only 28% of the increase, since part-time employees are not widely covered for these types of leave.¹⁶

In summary, the increase in hours lost for full-time workers seems largely attributable to the increased presence of older workers, who have more fringe benefits. Also, workers in general appear to be assigning more value to employment conditions that favour a better balance between work and personal life, since the combined increase in maternity leave and leave for personal or family responsibilities accounts for more than 40% of the overall increase in hours lost. A major proportion of part-time employees have already by definition struck this balance, since for most of them, working part time is a matter of choice.¹⁷

Other factors
A regression model was used to examine factors such as the increased proportion of employees working part time, aging, region of residence (province and urban/rural area), the temporary nature of the job, student status, and occupation and industry in order to test their effects on the decrease in average actual hours worked.

Between 2000 and 2003, the proportion of employees working part time went from 17.2% to 18.0%. However, the proportion dipped to 17.7% in 2004. The increase in part-time work was greater for men, with their proportion rising from 9.6% to 10.3% in 2003 and 10.2% in 2004. For women, it rose from 25.9% in 2000 to 26.5% in 2003, then declined to 26.0% in 2004 (Table 4).

This increased propensity for part-time work accounted for 20%¹⁸ of the drop in average hours, a sizeable share. Once again, it would be tempting to say that aging is a factor. For women, the increased propensity was more pronounced for the older age groups. For men, however, it was distributed among most age groups. The explanation could therefore be a lack of full-time jobs, since a larger share of young men take part-time work by necessity. It could also be that some of these young men are students, making it impossible to rule out the hypothesis of an increased preference for a better balance between work and personal life.

Variables such as occupation and industry do little to explain the decline in hours. However, what little these variables add would seem to show that a small part of the decline in hours is attributable to some transfer of jobs from occupations and industries with relatively high hours to ones requiring fewer hours.

Did the reference week bias have an effect in the past?

Since the degree of representation of statutory holidays in the reference week has a major impact on the trend in hours between 2000 and 2003, it is legitimate to ask whether such a bias occurred in the past. Hours actually worked were adjusted starting in 1987 (Chart D).

A comparison of adjusted and unadjusted hours shows that similar patterns occurred in the past. One of the most important took place in the early 1990s. In 1989 and 1990, few statutory holidays were captured by the LFS (see the second table in Definitions and the reference week).¹⁹ The adjustment of hours actually worked therefore had little effect during these years. Starting in 1990, the economic slowdown was apparent; hours (adjusted or otherwise) decreased substantially. In 1992, several statutory holidays were captured by the LFS (Good Friday, Thanksgiving and Remembrance Day).²⁰ The adjustment of actual hours therefore had more of an effect in 1992, significantly raising the number of hours worked. Unadjusted actual hours between 1990 and 1992 indicate a much larger drop than the adjusted hours.

Subsequently, from 1994 to 1999, adjusted and non-adjusted hours increased substantially and at the same pace. Starting in 2000, growth was slower, for both adjusted and non-adjusted hours. (For a description of the methodology used by the United States and the European Union, see Labour force surveys in other countries.)

Impact of hours on productivity

The reference week bias in hours actually worked affects not only the reading of labour market indicators but also labour productivity, since the latter reflects production per hour actually worked. Unless LFS hours are adjusted for various captured and non-captured holidays, the productivity measure would indicate spurious changes and would give an erroneous measure of economic efficiency. For this reason, before productivity is calculated, several adjustments are made to hours actually worked (see Adjustments made by the Canadian Productivity Accounts and Maynard 2004).

The adjusted hours are usually lower than non-adjusted hours (Table 5). The difference may amount to as much as 12.4 days annually (1989). Adjusted annual hours also fluctuate much less. For example, between 1987 and 2004, they ranged between -1.1% and 0.8% compared with -2.0% to 1.5% for unadjusted hours.

The effect of using unadjusted hours in calculating productivity was also measured. Taking unadjusted hours as the denominator, labour productivity in 2001 would be overestimated by 1.5%. This overestimation would amount to an average of 0.9% per year between 2000 and 2003 and 0.6% between 1989 and 1993. However, over the long term (1987 to 2004), the effect is marginal (0.1%).

Conclusion

Between 2000 and 2003, the LFS estimate of annual average hours of work gradually declined by 70 hours, or the equivalent of two weeks of work. This decline was surprising, since employment continued to be uncommonly strong despite slower economic growth than in the late 1990s. This strength contrasted with the stagnation of employment in the United States. Some analysts therefore questioned the strength of the labour market during these years.

More than half of the decrease was in fact due to survey methodology. The under-representation of some statutory holidays led to an overestimate of annual average hours worked in 2000 and thus, by comparison, an exaggerated drop in hours in the next three years. Similar patterns have occurred previously, notably between 1989 and 1992.

Once adjusted to eliminate the statutory holiday bias, the decline amounted to an average of one week annually per employee instead of two. Two-thirds of the decrease in adjusted hours came from an increase in hours lost for other than statutory holidays. This increase in hours lost was attributable to the aging of the workforce, since a major portion of the increase was seen among workers aged 45 and over.

However, aging was not the only factor. Increases in time off for vacation and for personal or family responsibilities, as well as changes to EI that resulted in more weeks of maternity, parental and adoption benefits as of December 31, 2000, also contributed, reflecting the greater value assigned to a better balance between work and personal life. The increased propensity to work part time, which was more pronounced among men in all age groups, could also be seen as reinforcing this trend.

In addition, 2003 was disrupted by several unfortunate events: the August power blackout in Ontario, concerns about a possible SARS epidemic, and forest fires and floods in British Columbia. Combined with the substantial appreciation of the Canadian dollar, these events led to an increase in work absences for other reasons, which accounted for nearly one-fifth of the total increase in hours lost.

Thus the decrease in adjusted hours did not seem to reflect a lack of economic vitality, but rather the aging of the workforce and the greater value assigned to a better balance between work and personal life.

Nonetheless, unless adjusted, the LFS estimate of hours actually worked often introduces a bias that can distort interpretation of labour market conditions. A comprehensive adjustment is produced regularly at Statistics Canada in the Canadian Productivity Accounts program. The Current Population Survey, the American counterpart of Canada's LFS, uses virtually identical procedures, so their estimates also contain a reference-week bias. On the other hand, new labour force surveys by nations in the European Union gather their data weekly from reduced samples, nullifying this bias.

Notes

1. This is the definition of usual hours used since January 1997. Prior to that, usual hours included any overtime hours usually worked by the survey respondent during a normal workweek, regardless of whether those hours were remunerated. Usual hours are used to calculate hourly wage rates.

2. Actual hours are used to calculate productivity and the hourly cost of labour.

3. In 2000, the system was new, and messages reminding respondents of the Thanksgiving and Remembrance Day holidays did not function. As a result, the LFS greatly underestimated the hours lost that year because of these holidays.

4. Self-employed workers report only absences of one week or more. Employees additionally report absences for part of a week. In this article, total hours lost cover both types of absence. Therefore, self-employed workers are excluded.

5. It is important to distinguish between average hours per employee and total hours. Total hours continued to grow between 2000 and 2003, since employment increased. However, they grew less rapidly, and hence the decrease in average hours.

6. Other reasons include time lost because of weather, strikes, lockouts, temporary layoffs, job starting or ending during the week, lack of material, or maintenance and repair of work premises.

7. Hours lost are annualized by multiplying weekly hours by 52 and by dividing the result by 7.5 hours per day.

8. From January to October, the LFS reference week includes the 15th day of the month. In December, the reference week is moved ahead to avoid having interviews take place during the weeks prior to Christmas. The same is usually done for November so that at least three weeks will separate the November and December interviews (David 1989).

9. Some holidays occur on a fixed date (Christmas, New Years Day, Remembrance Day) while others are variable (Civic Holiday, Good Friday). Some workers are penalized when fixed-date holidays fall on a weekend.

10. Not all workers are given this holiday, but most provincial and federal employees are. (See second table in Definitions and reference week.)

11. For further details, see Definitions and reference week.

12. The adjustment consisted of adding hours lost for statutory holidays to actual hours of work. An additional adjustment was made for 2000, when the LFS reference week did not take place during the vacation of construction employees in Quebec, whereas in 2003, those vacation weeks were captured. In this case, the hours lost to vacation are estimated on the basis of usual hours. The average actual hours thus adjusted are slightly overestimated, since hours lost to statutory holidays are considered as having been worked, whereas in realty, this is not the case. However, this adjustment eliminates the bias caused by the reference week, and the numbers of hours may therefore be compared without risk of error.

13. The years 2000 and 2003 represent the peak and trough respectively of average actual hours. The rest of the article examines differences between these two years to explain the decline in hours.

14. Only women can report hours of absence owing to maternity leave. Men can take parental leave, which is captured in leave for personal or family responsibilities.

15. Some laws have reinforced this effect, such as the Quebec legislation that gradually lowered the standard workweek from 44 hours in 1998 to 40 hours in 2002. A similar regulation in Ontario established a ceiling of 48 hours, including overtime.

16. An Oaxaca decomposition model was used to see whether the increase in hours lost was more concentrated in specific industries and occupations or the result of a transfer of employment from some industries and occupations with low levels of hours lost toward industries and occupations with high levels. However, the model showed that the increase was broadly based. The decomposition was done by estimating a linear regression model with the average number of hours of absence per week as the dependent variable and the following independent variables: age group, sex, temporary employee status, student status, province of residence, rural/urban area, occupation and industry.

17. The proportion of employees working part time by choice was more than 70% in 2000 and in 2003.

18. This estimate of 20% was obtained through a series of ordinary least squares regressions in which the dependent variable was an average of the number of adjusted actual hours for the years 2000 and 2003 combined. In the first regression, a single dependent variable was used, namely a dichotomous variable taking the value 1 in the case of data for 2003 and 0 otherwise. In a second regression, the proportion of part-time employees was added. The coefficient of the dichotomous variable then declined by 20%, meaning that the growing proportion of part-time employees explains 20% of the decrease in hours. The third regression included the variables in the second regression as well as province and the rural/urban nature of the area of residence. A fourth regression included all variables in the third regression as well as occupation and industry, the temporary or non-temporary nature of the job, and student status. The coefficient for Model 5 was -0.54, or 31% less than that of the first model, which shows that adding all the variables in models 3, 4 and 5 managed to add only about 10 percentage points more than Model 2 to explaining the drop in hours.

19. In 1989 and 1990, the only holiday captured was Remembrance Day. Also, this holiday fell on the weekend in both these years. When Remembrance Day falls on the weekend, it often has less impact on hours lost, since eligible employees would carry the holiday over to the following Monday.

20. In 1992, Remembrance Day fell on a weekday. Also, the July reference week took place during the vacation period for Quebec construction workers, reducing actual hours for that month.

References

• Akyeampong, Ernest. 2002. "Unionization and fringe benefits." Perspectives on Labour and Income (Statistics Canada, catalogue no. 75-001-XIE). August 2002 online edition.

• David, Suzanne. 1989. Total annual hours worked: Definitions and measurement approach. Staff report. Statistics Canada, Labour and Household Surveys Analysis Division.

• Fortin, Pierre. 2003. Différences dans les heures annuelles travaillées par habitant entre les États-Unis et le Canada. Université du Québec à Montréal. Brief presented in New York, April 16, 2003.

• Marshall, Katherine. 2003. "Benefiting from extended parental leave." Perspectives on Labour and Income (Statistics Canada, catalogue no. 75-001-XIE). March 2003 online edition.

• Maynard, Jean-Pierre. 2004. Annual measure of the volume of work consistent with the SNA: the Canadian experience. Paper presented to the Paris Group annual meeting in Lisbon, Portugal, September 29 to October 1, 2004.

• Royal Bank of Canada (RBC). 2004. "Canada's labour market: Looking good from far, but far from good?" August 4, 2004.

• Statistics Canada. 2005. The Canadian productivity accounts: data. Catalogue no. 15-003-XIE, vol. 1, no. 2 Ottawa.

Full article in PDF

Authors
Diane Galarneau is with the Labour and Household Surveys Analysis Division. She can be reached at (613) 951-4626, Jean-Pierre Maynard and Jin Lee are with the Micro-economic Studies and Analysis Division. Jean-Pierre Maynard can be reached at (613) 951-3654 and Jin Lee at (613) 951-1174. All authors can be reached at perspectives@statcan.gc.ca.

You need to use the free Adobe Reader to view PDF documents. To view (open) these files, simply click on the link. To download (save) them, right-click on the link. Note that if you are using Internet Explorer or AOL, PDF documents sometimes do not open properly. See Troubleshooting PDFs. PDF documents may not be accessible by some devices. For more information, visit the Adobe website or contact us for assistance.