...highlight relatively good fit of the model for the broad range of middle age groups and illustrate our concerns about the deviation of actual observations from the model predictions for the youngest and the oldest age groups.

[FIGURE 13 OMITTED]

For the 16-19 age group, the model appears to have captured the general trends and turning points in participation rates, although for teenage males there have been long stretches where the model prediction deviated from the data. Most recently, the model has expected participation rates to recover back toward their trends. In fact, actual rates have remained well below trend; although this finding represents a failure of the model, it does mitigate the concern that endpoint bias may be dragging down the estimated trend. Note, however, that for ages 20-24 the model exhibits much smaller errors, which suggests that the model residuals for the teenage groups have not tended to be carried through to older age categories as these cohorts age, and thus that the changes affecting recent cohorts of teens are age specific rather than cohort specific.

For ages 25-61 the model fits well overall and is not surprised by the developments of the past few years. Notably, the model effectively captures the dramatic change in slope in the participation rate of prime-age women and the persistent downtrend in the participation rate of prime-age men. In contrast, the actual participation rates for both older men and older women have exceeded the model predictions in recent years. The model also missed actual outcomes fairly uniformly across the older age groups in some earlier periods (for example, 1985-86), suggesting that we may have omitted some salient influence on retirement decisions from the model. For example, the errors in the most recent few years could be related to sizable movements in asset valuations, although, as noted above, we did not find variables representing wealth to be significant in the model. Nevertheless, the large and growing size of this group suggests that these errors represent a substantial risk to our estimated trend.

Model Projections and Alternative Simulations

We can also use the model to project how the trend in labor force participation will develop in coming years. To do this we employ the following procedure. For birth-year cohorts age 16 or above in 2005, we hold the cohort effects constant at their last values and essentially age these cohorts along the last observed age profile. (47) For cohorts who had not yet entered the labor market by 2005 (so that we have no model estimate of their cohort effect), we assume that the cohort effect is constant at the average value of the last few cohorts and then age them along the last observed age profile.

As figure 12 shows, the model projects that the trend in the aggregate labor force participation rate will fall further over the next ten years; indeed, the projected decline from 2005 to 2015 is more than 3 percentage points, which is comparable to the increase over the first ten years of our estimation period, when female participation was rising so rapidly. About 2 percentage points of this decline reflects the projected changes in the age distribution of the population associated with the aging of the baby-boom cohort, and the remainder is due to the model's estimates of the trends in the age and cohort profiles over the next ten years.

In constructing this projection of the trend, we assumed that the sizable recent model errors for teenagers and for the oldest age groups were not a manifestation of changes in the trend. However, an alternative approach would be to interpret the errors as suggestive of a recent change in the age profiles at those ages. To examine how this alternative interpretation would change our projection of the trend, we added the average error over the last two years to the age effects for teenagers and for the 62-and-over age group. For the latter group, this change reduces the extent of the drop in the age profile for older ages and, as indicated in the top panel of figure 12, raises the level of the projected trend by 1/4 percentage point by 2015. For teenagers this exercise steepens the age profile between youths and prime-age individuals and reduces the aggregate trend by 1/4 percentage point. Hence, as it happens, carrying forward both sets of errors leaves the projected trend almost unaltered.

Alternatively, the recent errors in the teenage equations could indicate that the attachment of recent cohorts has fallen more sharply than allowed for in the baseline model. To simulate this possibility, we added the average recent errors among teenagers to their cohort effects and computed the effects on the aggregate trend as they age. This simulation, shown by the lower line in the bottom panel of figure 12, leads to a steeper projected decline in the aggregate trend and reduces the trend level by more than 1 percentage point in 2015. Of course, it also seems possible that these cohorts might eventually have greater labor force attachment than the model suggests. For example, if the additional schooling obtained by these cohorts boosts their participation rates throughout their lives, the baseline model would underestimate the aggregate trend. We approximated this influence by raising the cohort effects for cohorts born after 1984 to that of the 1984 birth cohort, rather than allowing any further decline (in effect raising the average participation rates of these cohorts later in life). This simulation, shown by the upper line in the figure, slows the decline in the trend noticeably and produces a trend level that is about 1/2 percentage point higher in 2015 than that projected by the baseline model.

We also recognize that a steeper rise in the labor supply of older workers than predicted by the model is a realistic possibility, especially if the aging of the baby-boom cohort leads to changes to parameters of the Social Security program or to concerns about the viability of private pension plans or retiree health benefits. However, given the substantial downward pressure on the aggregate participation rate from other forces, participation rates among this group would need to rise dramatically to prevent future declines in the aggregate trend participation rate. For example, if participation among the remaining age groups turns out as the model predicts, the average participation rate of individuals 62 and over would need to double over the next ten years, from 20 percent to 40 percent, to hold the trend at its current level. Such a change would require a quickening of the pace of increase in this group's participation rate from roughly 0.1 percentage point a year recently to 2 percentage points a year. (48) Given the projected increases in the number of individuals in this group who are over 80 years of age (and therefore unlikely to work), such a sharp acceleration seems unlikely.

Additional Evidence

Although the results are not directly comparable with those from the cohort-based model presented above, other aspects of recent patterns in labor force participation can provide independent evidence on the extent to which changes in the aggregate labor force participation rate in recent years are cyclical or structural in nature. Here we present several such related analyses, including a comparison of participation rate changes in different states, an examination of gross labor force flows, and changes in the duration and incidence of labor force participation.

Cross-State Evidence

Variation in participation rates across states is one alternative source of information about the potential sources of the post-2000 decline in the aggregate participation rate. In particular, if changes in participation during this period were driven largely by changing labor demand conditions, one would expect those states in which the labor market showed a relatively greater deterioration to also have experienced a larger relative drop in labor force participation rates. On the other hand, to the extent that the changes in participation were unrelated to fluctuations in labor demand, one would expect them to be uncorrelated with a state's cyclical condition.

To investigate this proposition, we regressed the annual participation rate in each state on a constant state-specific effect, a common linear trend as a measure of structural factors, state-specific cyclical conditions, and a dummy variable equal to zero before 1994 and one otherwise, to control for any effects of the CPS redesign. To capture possible changes both in the underlying trend rate of participation and in the responsiveness of the participation rate to the business cycle, we allow for a break in the coefficients on the trend and cycle terms after 2000. (49) The full specification is as follows:

(5) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where s indexes states, t indexes time, lfpr is the participation rate, cyc is the state unemployment rate (our measure of cyclical conditions), d00 is a dummy variable equal to one beginning in 2001 and zero before that, and d94 is the CPS redesign dummy. To control for spurious correlation between the unemployment rate and the participation rate due to measurement error, we instrument for the contemporaneous unemployment rate with a state's lagged unemployment rate and the contemporaneous percentage change in payroll employment. The model is estimated using population-weighted least squares and data from 1990 to 2005. The estimated coefficients (except for the state effects) are reported below each parameter, with t statistics shown in parentheses.

If the post-2000 downward movements in the participation rate were associated with structural factors uncorrelated with changes in state-level labor demand, our estimates of [[delta].sup.t] should be negative. If, on the other hand, participation rate declines were caused only by changes in demand, with or without an increase in the cyclical sensitivity of the participation rate, our estimates of [[delta].sup.t] should be zero and our estimates of [beta] should be negative. If changes in cyclical sensitivity played an important role in the post-2000 behavior of the participation rate, our estimates of [[delta].sup.c] should be negative.

Overall, the estimation results suggest that both cyclical and structural factors played a role in the post-2000 decline. Estimates of 8' and [3 are both negative and statistically significant. On net, the point estimates imply that the break in the common trend accounts for about one-half of the 0.8-percentage-point decline in the participation rate between 2000 and 2005, with the remainder accounted for by changes in cyclical conditions. (50) Although the estimate of [[infinity].sup.c] is also negative, it is not statistically significant at conventional levels, suggesting little or no change in the cyclical sensitivity of the participation rate. Despite the differences in the information used to identify structural changes, these results are quite similar to those from our cohort-based model, which also estimates that about half of the decline in the participation rate since 2000 was due to structural forces.

Gross Labor Force Flows

Patterns of gross labor force flows may also be useful in discerning the reasons for the post-2000 drop in the participation rate, given a set of assumptions about the types of flows that would be expected to be associated with cyclical and structural changes in participation. In particular, one reasonable presumption is that withdrawal from the labor force as an unusually strong response to the weak job market in recent years should be reflected in an unusually large rate of flows out of unemployment into nonparticipation, as job seekers became discouraged. In contrast, the flow out of employment into nonparticipation arguably should be procyclical, because employed individuals, to the extent that they are worried about job prospects, would be reluctant to leave the labor force temporarily (for example, to go back to school) in a weak economy. As a result, any increase in this latter flow during and after the 2001 recession would likely be related to more structural factors.

Figure 14 shows the rates of flow out of employment and unemployment into nonparticipation. (51) As expected, the flow rate from unemployment to nonparticipation tends to increase when the job market weakens, whereas that from employment to nonparticipation tends to decrease. This evidence suggests that we can use the deviations from these standard cyclical patterns as a test of whether the post-2000 decline in participation was due to structural factors or to an unusually strong response to the cyclical deterioration in the labor market.

[FIGURE 14 OMITTED]

To implement this test, we first estimate the pre-2001 typical cyclical response using the following equations:

(6) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where un is the rate of flow from unemployment to nonparticipation, en is the rate of flow from employment to nonparticipation, cyc is a measure of the stage of the business cycle (we use the unemployment rate), d94 is a dummy variable equal to one in 1994 and later and zero before 1994 (to control for the CPS redesign), and t and [t.sup.2] are linear and quadratic trend terms, respectively. Coefficient estimates and t statistics are reported below each parameter. Using the estimated [beta]s, we then construct estimates of the flow rates, excluding cyclical effects, through 2005 as

(7) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

Lastly, we regress these cyclically adjusted measures on quadratic time trends and a dummy variable (dO0) set equal to zero before 2001 and to one thereafter:

(8) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

This specification embodies our assessment that, through 2000, the underlying trend in both flow rates can be reasonably well described by a quadratic time trend. After 2000, however, we allow the average flow rates (excluding the typical cyclical response) to be freely estimated. In this way the average post-2000 fitted values will reflect both the presence of excessive cyclical responses and underlying structural change. Under our assumptions, a higher rate of flow from unemployment to nonparticipation (after controlling for the typical cyclical response) would support the excess cyclicality hypothesis, whereas a higher rate of flow from employment to nonparticipation would favor the structural change hypothesis.

Estimation results support the latter hypothesis. The mean value of the post-2000 flow rates of unemployment to nonparticipation (excluding the typical cyclical response) is slightly lower than the average flow rate from 1994-2000, but this difference is not statistically significant. In contrast, the difference between the mean post-2000 employment-to-nonparticipation flow rate and the average pre-2001 flow rate is substantial and positive, and this difference is statistically significant.

The magnitudes of the changes in cyclical and structural flows imply that all of the change in the participation rate since 2000...

NOTE: All illustrations and photos have been removed from this article.




Looking for additional articles?
Search our database of over 3 million articles.

Looking for more in-depth information on this industry?
Search our complete database of Industry & Market reports by text, subject, publication name or publication date.

About Goliath
Whether you're looking for sales prospects, competitive information, company analysis or best practices in managing your organization, Goliath can help you meet your business needs.

Our extensive business information databases empower business professionals with both the breadth and depth of credible, authoritative information they need to support their business goals. Whether it be strategic planning, sales prospecting, company research or defining management best practices - Goliath is your leading source for accurate information.