Disability insurance denial rates and the labor force participation of older men and women in...

Introduction

A large expansion in the number of individuals collecting disability insurance benefits has been observed in many industrialized countries. This growth, especially in the United States, has led some researchers to examine the causes of these increases. Most of this research

has focused on the disincentive effects of disability benefits on labor supply. (1) By comparison, relatively little research has examined other features of disability policy such as the criteria used to determine the eligibility for benefits, which can be summarized in the denial rates for applicants. Changes in these criteria can also affect an individual's application behavior and labor supply decisions. For example, Gruber and Kubik (1997) examined the effect of rejection rates from the US Disability Insurance (DI) program on the labor force participation of older men and found that a 10 percent rise in denial rates was associated with a statistically significant 2.8 percent decline in labor force nonparticipation, among men aged 45-64. More recently, Campolieti (2003) examined the effect of the 1989 changes in the Canada Pension Plan disability program's eligibility criteria, which made the screening of applications less stringent, on the labor force participation of older men. Using a difference-in-difference estimator, with standard errors that properly accounted for the clustering in the data, and data from 1987 to 1991, Campolieti found that these changes in eligibility criteria were associated with a 1.5 percentage point increase in nonparticipation. (2)

The disability literature has also, for the most part, focused almost exclusively on older men. This is unfortunate because older women have become a fairly large proportion of the beneficiaries on the disability rolls. For example, in Canada older women (between the ages of 45 and 64) increased from 12.8 to 35.6 percent of all beneficiaries on the disability rolls between 1971 and 1997 (Campolieti, 2001). This acceleration in the number of female applicants became much more prevalent during the late 1980s and early 1990s, when the incidence rate for new female beneficiaries increased from 2.99 per thousand persons in 1989 to 5.79 per thousand persons in 1994, while the incidence rate for men grew from 4.28 to 6.23 per thousand persons during the same period (Canada Pension Plan Consultations, 1996). Given these trends, examining the effects of disability programs on the labor force behavior of older women may be very important from the perspective of policy makers.

In this paper, we use data from Canada to examine the effect of denial

rates for disability insurance benefits on the labor force participation of older men and women. In Canada, disability insurance benefits are provided by the Canada/Quebec Pension Plan (C/QPP) disability program. This structure means that the QPP program only operates in the province of Quebec and the CPP program covers persons in the rest of Canada. Since these programs are administered separately there will be variation in administrative procedures between programs, i.e., the denial rate for applicants. In addition, after 1995 applications to the CPP disability program were adjudicated at regional offices, which interpreted the policy directives that the program administrators in the national headquarters issued. We use this administrative setup to exploit a number of changes in eligibility criteria and medical screening occurring during the mid 1990s that provide both time series and cross-sectional variation in the denial rates between the CPP and QPP programs as well as within the CPP program.

The less stringent medical screening in the CPP disability program prior to the changes in administrative procedures could have created moral hazard problems as able-bodied individuals applied and were occasionally granted disability benefits. The variation in the denial rates created by this change in policy and the structure of the initial screening system will permit us to examine the effect of these more stringent screening criteria on the labor force participation decisions of older persons. We obtain our estimates using an estimator suggested by Wooldridge (2003) that properly accounts for the clustering in the data when computing the standard errors. This clustering in the data arises because individual level employment outcomes are regressed on policy variables (i.e., denial rates) that only vary across groups. Not adjusting the standard errors for this clustering will produce standard errors that are too small (Donald & Lang, 2001; Moulton, 1990; Wooldridge, 2003).

This paper makes a number of contributions to the literature. First, we will examine the effect of denial rates on the participation decisions of older men following a change in screening stringency in Canada. Second, we will estimate the effect Of an increase in screening stringency on the labor supply of older women, which has not been examined before in Canada or the United States. Third, unlike the other papers in this literature, we compute the appropriate standard errors on the estimates capturing the effect of denial rates on the labor force participation decisions of older persons. All of our estimates from our preferred specifications suggest that denial rates do not have the hypothesized statistically significant negative effect on the participation decisions of older men and women. Finally, we will discuss how changes in the CPP applicant pool might explain our results. One implication of our findings is that it may be more appropriate to obtain a menu of alternative estimates that correspond to different policy regimes rather than presenting a single estimate. We also discuss a few additional policy implications of our findings for other programs that support disabled persons and some suggestions for future research.

The Canada/Quebec Pension Plan Disability Program

The Canadian analogue to the US Social Security Administration's (SSA) Disability Insurance (DI) program is the Canada/Quebec Pension Plan disability program. The most important difference between disability insurance arrangements in Canada and the US is that the Canadian system contains two programs: the Quebec Pension Plan (QPP) disability program, which only operates in Quebec, and the Canada Pension Plan (CPP) disability program, which operates in the rest of Canada. Individuals are eligible for C/QPP disability benefits if they suffer from a prolonged mental or physical condition that prevents them from working as well as satisfying a contribution requirement, which provides a measure of attachment to the labor force.

Our interest centers on a number of changes in eligibility and screening requirements that occurred in the CPP disability program, while the QPP disability program remained unchanged. The CPP began a series of administrative changes in the middle of 1994 to the disability program's applications review procedures to deal with its increasing caseload. Following these changes a more stringent set of medical adjudication guidelines were adopted in September 1995. As well, after September 1995 socioeconomic and economic factors could no longer be used as mitigating factors and those aged 55-64 faced the same work test as other applicants. (3)

We computed rejection or denial rates for applicants at the initial stage, excluding applicants that did not satisfy the contribution requirements to the plans, for both the CPP and QPP disability programs and present them in Table 1. The rejection rates for the QPP disability program have been fairly constant during this period. However, there has been a significant increase in the denial rate for the CPP disability program beginning in 1994, which reached a peak in 1997 before stabilizing at a slightly lower level.

The preceding discussion suggests that the changes in the administrative handling of applications in the CPP disability program provided variation in the denial rates for disability benefits. Gruber and Kubik (1997) examined a similar increase in denial rates in the United States. In the United States, the Social Security Administration (SSA) provides the DI program. However, applications to the DI program are handled by state level boards, which can interpret the SSA guidelines. Gruber and Kubik used an increase in denial rates that occurred in the late 1970s as the SSA moved to deal with the increasing caseload in the DI program by implementing more stringent adjudication guidelines. The increase in screening stringency created time series variation in denial rates. Gruber and Kubik also argued that the state level screening boards for DI benefits created cross-sectional variation in denial rates because these adjudication boards differed in their interpretation of the new SSA guidelines. As a result, some states had large increases in denial rates, while others had relatively small increases. Gruber and Kubik then examined the differential response of this policy change on the labor force participation decisions of older men. In particular, were the states that increased the denial rate for DI benefits the most the ones that experienced the largest increases in the labor force participation of older men?

The CPP disability program administers applications for benefits in a similar fashion to the SSA. Prior to 1996, applications to the CPP disability program were handled at the national level at Human Resources Development Canada (HRDC). However, in response to a number of critiques made by the Auditor General of Canada, HRDC began handling the adjudication of CPP disability applications in regional offices in 1996. (4) The HRDC policy directive that required decisions for CPP disability benefits to be based entirely on medical reasons and give no consideration to local labor market or economic conditions after September 1995 (Human Resources Development Canada, 1996) would have created time series variation in CPP denial rates. Some evidence to illustrate this can be found in an administrative study conducted by HRDC in which the new adjudication standards that were implemented in 1995, which were based exclusively on medical grounds, were applied to a group of files that had decisions based on the old CPP guidelines (Human Resources Development Canada, 1996). The results from this evaluation exercise suggested that 11 percent of the grants made under the old HRDC guidelines would have been rejected and additional information would be needed on another 21 percent of the files before reaching a final decision (Human Resources Development Canada, 1996). This indicates that given the same applicant pool the new adjudication requirements would have produced a higher denial rate for applicants.

Since the regional offices that adjudicated the claims beginning in 1996 were able to interpret HRDC's new guidelines, the regional offices could differ in their interpretation of the national guidelines set by HRDC. This discretion at the regional level led to provincial differences in the denial rates for CPP disability benefits as some provinces had high denial rates and others smaller denial rates (Torjman, 2002).

The changes in the administration of CPP disability applications created both time series and cross-sectional variation that is like that exploited by US studies. The administrative setup of the CPP disability programs and the variation in the denial rates makes it possible to determine the effects of these denial rates on the participation decisions of older persons. In particular, we can examine the differential response of the changes in denial rates on the labor force non-participation of older persons in Canada.

In addition, the QPP disability program had its own adjudication standards that did not vary during the study period (i.e., there were no policy directives regarding changes in adjudication standards), which differed from those used in the CPP disability program. In particular, the QPP program tended to be tougher on musculoskeletal and soft-tissue conditions as well as mental disorders than the CPP program. HRDC also conducted another administrative review of CPP grants in which QPP adjudicators reviewed CPP files using the QPP guidelines (Human Resources Development Canada (1996)). The QPP adjudicators rejected 30 percent of the CPP grants for those with musculoskeletal conditions and would have required additional information on another 44 percent of the files before reaching a decision based on the QPP guidelines. Similarly, for mental disorders the QPP adjudicators would have rejected 9 percent of the files and require additional information on another 60 percent of the CPP decisions before making their determination. This is a source of cross-sectional variation that US studies are not able to exploit because the United States does not have separate disability programs for different regions.

Empirical Strategy

We use a two-step procedure (detailed subsequently) to estimate linear probability models of the form

NP = [alpha] + x' [beta] + [[delta].sub.1] REJ + [[delta].sub.2] DISAB + [phi]' [theta] + [tau]' [lambda] + u, (1)

where NP is a dummy variable for nonparticipation in the labor force, x is a vector containing controls for age, education and marital status, REJ is the denial rate from the initial stage of the C/QPP disability program application process, DISAB is a dummy variable that indicates the individual is in poor health, [phi] is a vector of province dummies, [tau] is a set of year dummies and u is a residual term. The province dummies capture systematic fixed differences across provinces in the nature of the applicant pool, while the year dummies capture time series differences in the trends of labor force participation decisions. Increases in denial rates should decrease labor force nonparticipation, so we expect [[delta].sub.1] to have a negative estimate. This is the principal hypothesis that is being tested in this paper.

We control for disability status using a variable based on a body mass index (BMI), which is computed as weight (in kilograms) over height (in meters) squared. A 'healthy' value of the BMI for an adult aged 45-64 is between 22 and 28 (Kushner, 1993). Values outside the upper bound of this range, i.e., the obese range, are associated with greater risks of cardiovascular diseases, diabetes and colon cancer. Conversely, values less than the lower bound, i.e., the underweight range, are associated with greater risks of respiratory diseases. We created a dummy variable that indicated whether the individual had a BMI outside the healthy range.

We estimate Eq. 1 using a procedure suggested by Wooldridge (2003), which is equivalent to the methods used in Loeb and Bound (1996) and Baker and Fortin (2001). As has been noted in the literature, regressing individual level outcomes (e.g., whether the individual does not participate in the labor force) on variables that only vary across groups (e.g., REJ) can result in downward biased standard errors for the grouped variables (Moulton, 1990). Since the principal variable of interest in this analysis is the denial rate, which only varies across adjudication regions (whether within the CPP program or across the CPP and QPP programs), accounting for the clustering in the data to obtain the proper standard error is important. We selected Woolridge's procedure primarily because it relatively easy to implement.

Wooldridge proposed an efficient minimum distance (minimum chi-squared) estimator to obtain the standard errors that adjust for the clustering in the data. His procedure has two steps. The first step requires estimating (with ordinary least squares) the individual level outcome on variables that vary across individuals and a set of group-specific intercepts

[NP.sub.ig] = [x'.sub.ig][beta] + [[delta.sub.2] [DISAB.sub.ig] + [Group'.sub.ig] [psi] + [v.sub.ig] (2)

where i indexes the individual and g the groups (i.e., adjudication regions), [Group.sub.ig] is a set of group-specific dummy variables that correspond to the different adjudication regions in Canada during our study period and [psi] is a G x 1 vector. The second step of the procedure regresses the estimates of the group-specific intercepts (i.e., the [[??].sub.g]s from the first-step) on the K variables that only vary across groups

[[??].sub.g] = [[alpha].sub.0] + [[alpha].sub.1] [REJ.sub.g] + [[phi]'.sub.g][eta] + [[tau]'.sub.g][omega] + [e.sub.g], for g = 1, ..., G. (3)

Equation 3 is estimated using weighted least squares, with the weights given by the asymptotic variances, which were adjusted for heteroskedasticity with White's procedure, on the group-specific intercepts from the first-step. The standard errors on the grouped variables in the second-step regression will properly account for the clustering in the data. Wooldridge (2003) also proposed using the weighted sum of squared residuals from the second stage regression as a check for the fit of the model. This test statistic, which he refers to as an over-identification test statistic, will have a chi-squared distribution with G-K-1 degrees of freedom, where G is the number of groups in the data and K is the number of explanatory variables in the second stage regression.

Data

We use data from Statistics Canada's National Population Health Survey (NPHS) to estimate our models. The NPHS was administered for the first time in 1994 and 1995 (Wave 1), but there have been additional surveys conducted in 1996 and 1997 (Wave 2) as well as 1998 and 1999 (Wave 3). The survey is collected throughout the first year and in the first quarter of the second year of each wave. The NPHS contains detailed health information as well as some socioeconomic and labor market information.

We matched denial rates for the initial stage of the application process for C/QPP disability benefits from 1994 to 1995 (our 'before' period) as well as 1996-97 and 1998-99 (our 'after' period) to the NPHS. (5) Prior to 1996, we only have program level rejection rates. However, we are able to use the provincial level data on rejection rates in the CPP program beginning in 1996. (6) The first wave of the NPHS was collected before the September 1995 administrative directive was issued. Unfortunately, some other changes in the handling of applications also occurred during mid-1994. As a result, this limits our ability to fully exploit the 'before' and 'after' differences in the CPP denial rates with the NPHS data. However, the QPP disability program had stable adjudication procedures during the years the changes occurred in the rest of Canada. During the NPHS study period, 1994-1999, the denial rate in the CPP program increased by 23.7 percent, while those in the QPP program increased by 6 percent (see Table 1).

Our samples will include males and females between the ages of 45 and 64. The dependent variable is a dummy variable for nonparticipation in the labor force. The principal explanatory variable will be the denial rate from the first stage of the C/ QPP application process. We also included controls for age (dummy variables for ages 45-49, 50-54, 55-59, and 60-64), education (no high school, attended high school, some post secondary, completed university) and marital status. We control for disability status using a dummy variable that takes the value 1 if the individual had a BMI that was less than 22 or greater than 28 and 0 otherwise. A value of 1 indicates that an individual is at risk of poor health. We provide some summary statistics for these variables in Table 2.

The disadvantage of the NPHS is that the first wave was collected during a period when other changes in the handling of applications were occurring prior to the policy directive of September 1995. To examine the robustness of our findings to this we also use data from the Survey of Consumer Finances (SCF) to obtain our estimates. The SCF was an annual supplement to Statistic Canada's April Labour Force Survey that contains detailed information on labor market outcomes, educational attainment and demographic factors. We took 1992 and 1993 as our 'before' period and 1996 and 1997 as our 'after' period and matched the average denial rate for these periods to the data. We stopped the 'after' period in 1997 because that was the last year the SCF was collected. Using 1992 and 1993 as the before period means that we can avoid the period during 1994 when other changes in the handling of applications were made.

Unfortunately, the SCF does not contain any information on health status of any sort, which means that we cannot control for the effect of health status on labor force participation. However, the 'before' and 'after' design for the SCF data does provide much more variation in the denial rates than the NPHS study period. Specifically, during the SCF study period the denial rate increased by 43.6 percent in the CPP disability program, but decreased by 2.3 percent in the QPP program (see Table 1). This increased variation in the denial rates may make it easier to identify their effect on participation decisions. Sample statistics for the SCF data are also available in Table 2. Note that the SCF sample sizes (37,422 for males and 38,150 for females) are larger than those for the NPHS (12,832 males and 14,084 females).

Empirical Results

In order to conserve space, we only present and discuss the estimates for the denial rate, which is the principal explanatory variable of interest. The first-stage regression results for Wooldridge's procedure are available upon request. We estimate our models using two samples. The first sample includes both the within program variation in denial rates in the CPP program as well as the variation in denial rates between the CPP and QPP disability programs (i.e., observations from the CPP provinces as well as Quebec, which is covered by the QPP program). The second sample only includes the within program variation in CPP denial rates (i.e., no observations from Quebec). We also present the estimates including and excluding the year and province dummies. However, our preferred specification includes both the province and year dummies. This preference is dictated by two factors. First, the values of the Wooldridge overidentification test statistic clearly favor the specifications with the year and province dummies relative to those that exclude these variables. Second, the specification with the year and province dummies is most similar to what other researchers have used.

When the province and year dummies are excluded from the specification, the estimates on the denial rate for the older males based on the NPHS data, in Table 3, suggest that increases in denial rates for disability benefits are associated with statistically significant declines in labor force nonparticipation when the observations from the QPP disability program are included in the sample. When the QPP program is excluded from the NPHS sample we do not obtain a statistically significant estimate on the denial rate from this specification. However, when we estimated our preferred specification, which includes province and year dummy variables, the estimates on the denial rate differ a great deal. Both estimates (including and excluding the QPP disability program) are positive and statistically significant. The magnitude of the estimates between the NPHS sample including the QPP program and excluding the QPP program is also similar. Our results are consistent with the labor-leisure model where a reduction in non-wage income implies a new supply of labor at the existing market wage (if leisure is a normal good).

The coefficient estimates on the denial rates for older women based on the NPHS data are presented at the bottom of Table 3. Like the NPHS results for older men, excluding the year and province dummies produces a negative and statistically significant estimate on the denial rate when we include the observations on the QPP disability program in the sample, but not when we exclude them. When we added the year and province dummies to these specifications we do not obtain statistically significant estimates on the denial rate for older women's labor force participation decisions in either NPHS sample.

We also estimated some additional specifications with the prime age unemployment rate for men (aged 25-44) as a control for economic activity. These estimates are in line with the other estimates for the NPHS data in Table 3, although the magnitudes differ slightly.

The disadvantage of the analysis using the NPHS is that some other changes in the handling of CPP applications were implemented in 1994 prior to the change in screening requirements in September 1995. In order to examine the sensitivity of the estimates on the denial rate to this, we re-estimated our models with data from the SCF, which uses 1992 and 1993 as a 'before' period. However, as noted earlier, the SCF does not contain any health information, so we could not include a control for health status in the first-step of Wooldridge's procedure.

In Table 4, the denial rate does not have a statistically significant effect on the labor force nonparticipation decisions of older males using data from the SCF. This conclusion does not depend on whether we exclude the year and province dummies from the specification or the observations on the QPP disability program from the SCF sample. Similarly, the estimates for older women at the bottom of Table 4 also indicate that the denial rate does not have a statistically significant effect on labor force nonparticipation decisions. The estimates for older women are also not sensitive to whether we exclude the year and province dummies from the specification or the observations on the QPP disability program from the sample. The estimates for the SCF sample were also not sensitive to including the control for economic activity (the prime age male unemployment rate).

The results from our preferred specifications for the NPHS data suggest that denial rates do not have a statistically significant effect on the labor force nonparticipation decisions of older women. However, our estimates for older men suggest a positive and statistically significant relationship between denial rates and labor force participation decisions. Our results for older men and women from our preferred specifications based on SCF data indicate denial rates do not have a statistically significant effect on labor force nonparticipation. These estimates differ from the findings in Gruber and Kubik (1997) who found the hypothesized negative relationship between denial rates and nonparticipation for older men using U.S. data and Campolieti (2003) who found that less stringent screening requirements were associated with an increase in the labor force nonparticipation of older men in Canada. Our estimates are thus not consistent with previous studies and the predictions of economic theory. However, (except for Campolieti (2003)) these previous studies did not compute the appropriate standard errors to account for the presence of the grouped variables with micro data, unlike the estimates in this paper. Consequently, some of these existing estimates may not be statistically different from much smaller estimates if the appropriate standard errors, which adjust for the clustering in the data, are computed. For example, when we do not make any adjustment to the standard errors for the coefficient estimates on the denial rate they tend to be 8-40 percent smaller than the standard errors that are properly adjusted. Similarly, Campolieti (2004) has previously illustrated this difference in the size of the standard errors (cluster adjusted versus unadjusted) for disability benefit elasticities, so there could also be parallels in this empirical setting. There are also a number of plausible explanations for these differences, which are presented in the next section.

Discussion

Gruber and Kubik (1997) noted that a change in the applicantpool for disability benefits would bias the results against finding a negative correlation between denial rates and nonparticipation decisions. There are some interesting trends in the CPP and QPP disability programs during our study period that might reflect a change in the applicant pool. In the CPP disability program there was a decline in applications after the change in adjudication requirements (falling from 91,034 in 1995 to 56,121 in 2000). In contrast, the applications to the QPP program were relatively constant (i.e., no pronounced increases or decreases) during the same period. Since most of the conditions that disability beneficiaries report are chronic ones that result from cumulative lifetime exposure (Frank & Maetzel, 2000), it is unlikely that there could have been an 'exogenous' change in the composition of the medical conditions in the applicant pool during our study period. However, we believe there are a number of ways in which the change in the screening of CPP disability applications could have led to changes in the pool of persons applying for these benefits that could explain our results.

One plausible source of these changes might arise from the interactions of other social insurance programs on applications to the C/QPP disability program. Specifically, private disability insurers, provincial welfare and workers' compensation programs in Canada often view themselves as the 'residual' insurer and the CPP disability program as the 'first payer' and as a result they may encourage their beneficiaries to apply for CPP disability benefits (Human Resources Development Canada, 1996). This shifting between programs is less likely to occur in Quebec because the social insurance system there is more centralized and some legislation exists to prevent this from occurring (Campolieti & Lavis, 2000). Consequently, programs outside of Quebec often encourage their beneficiaries to apply for CPP disability benefits in order to reduce their own benefit expenditures (Human Resources Development Canada, 1996). The province and year dummies might be capturing cross-sectional and time series differences in these policies. This might explain why we obtained the anticipated effect on the denial rate when we excluded these dummies in our analysis of the NPHS data. More importantly, this variation in how different programs from different provinces view their role relative to the CPP disability program may influence the pool of CPP disability applicants. For example, some survey evidence indicates that as many as 17 percent of all CPP disability beneficiaries also receive benefits from a workers' compensation board (Human Resources Development Canada, 1996). As another example, the Ontario Ministry of Social Services prepared about 16,000 CPP disability applications for its welfare beneficiaries during 1993 and 1994 (Human Resources Development Canada, 1996). (7) Similarly, British Columbia and New Brunswick also undertook special initiatives in which all welfare recipients who were potentially able to apply for CPP disability benefits applied for those benefits (Torjman, 2002). This shifting of beneficiaries between programs might have become less common after the new adjudication requirements were in place after 1995. For example, only 3,863 CPP disability applications from Ontario welfare beneficiaries were processed in 1996 and 1997. (8) This represents a fairly large drop in the applications volume from 1993 and 1994, when 16,000 applications from welfare beneficiaries in Ontario were processed.

Alternatively, since the CPP disability program allowed economic conditions and socioeconomic factors (during the late 1980s-mid 1990s) to be taken into consideration when reviewing applications for disability benefits, the CPP disability program might have been used as a long-term unemployment insurance program. As a result, many applicants may have been able to work but unable to find work. One mechanism through which this moral hazard might have occurred is by more reporting of musculoskeletal and soft tissue conditions such as back injuries, which are the most prevalent medical diagnosis on the CPP disability rolls (Campolieti, 2002a). These conditions are known to present diagnostic and therapeutic challenges to health care providers (Agency for Health Care Policy and Research, 1994; Quebec Task Force on Spinal Disorders, 1987). A number of papers in the workers' compensation literature have found that workers are more likely to simulate work injuries with these conditions, which are hard-to-diagnose and verify (e.g., among others, Bolduc, Fortin, Labrecque, & Lanoie, 2002; Ruser, 1998). In related work, Campolieti (2002b) fbund that disability insurance denial rates were associated with a statistically significant decrease in the reports of back problems by older males, but had no effect on the reports of easier to diagnose chronic conditions. The increase in screening stringency could have increased the cost of reporting hard-to-diagnose conditions relative to an easy-diagnose conditions and led to changes in the applicant pool. Specifically, there could have been less scope for moral hazard on the part of workers reporting hard-to-diagnose conditions to apply for disability benefits.

Finally, the increase in the denial rates could have led to greater self-screening by applicants (Parsons, 1991). Since decisions on applications can be delayed some individuals might choose not to apply because there might be more uncertainty about the decision after the changes in adjudication requirements. Using aggregate state level US data, Parsons (1991) found that a 10 percent increase in denial rates induces a 4 percent decrease in applications after 2 years. However, Parsons's elasticities would only explain part of the decrease in applications to the CPP disability program, which suggests that the other factors we discussed might have contributed to the decline in applications.

Most of the discussion in this section suggests that it might not be possible to obtain a definitive estimate of the effect of denial rates on labor force participation decisions because other differences in policy might affect the relationship between denial rates and labor supply decisions. This means that rather than obtaining a single estimate what may be needed is a menu of alternative estimates that correspond to different policy regimes. Campolieti (2004) presented a similar argument when examining the effect of disability benefits on labor force participation decisions. In particular, Campolieti (2004) argued that it might not be possible to obtain a definitive estimate of the disability benefit elasticity because changes in other features of disability policy could affect the response of individuals. The same phenomena could explain the estimates for the changes in screening stringency obtained in this paper.

Another implication of the findings in this paper for other disability programs is that the individuals who are denied access to C/QPP disability benefits may have managed to find access to an alternative source of income, instead of labor market income. This means that another program in the social insurance safety net could have been utilized. However, it is not clear whether this is also true in Quebec, where the support system is more integrated and, consequently, prevents some overlap between different programs. Unfortunately, the information in our data is not rich enough to permit an exploration of these issues because it does not contain good income information with sufficient detail on sources of income. However, this would be an important question for future research to examine.

Finally, the changes in denial rates also implicitly mean that there are costs (as well as benefits) to private insurers and other disability programs because they would have to pay (not have to pay) for some individuals who could be shared with the C/QPP disability program. For example, if denial rates reflect less stringent screening criteria the C/QPP disability program would be supporting individuals that should only be receiving assistance from other programs. On the other hand, if eligibility criteria are too stringent it is possible that the other disability programs may not be able to effectively share costs with the C/QPP disability program. Consequently, there may be an optimal denial rate that could equate the marginal benefits and costs associated with more stringent screening criteria. Finding this socially optimal denial rate is beyond the scope of this paper, but it would be an important issue for future research to address.

Concluding Remarks

We examined the effects of an increase in denial rates from the C/QPP disability program on the labor force participation of older men and women in Canada. We computed the appropriate standard errors for the coefficient estimates on the denial rate using a procedure suggested by Wooldridge (2003), unlike earlier work. We did not find a statistically significant negative relationship (the anticipated and hypothesized effect) between denial rates and the labor force nonparticipation of older men and women when we estimated our preferred specification. This finding is consistent across both data sets we used as well as if we include or exclude the QPP program from our sample.

These findings might be explained by changes in the applicant pool for CPP disability benefits, which might lead to a positive relationship between the denial rate and labor force nonparticipation. However, the change in the applicant pool might have been related to a change in behavior of individuals as well as the administrators of other private and social insurance programs following the changes in screening stringency by the CPP disability program. In particular, there could have been fewer referrals from provincial workers' compensation or welfare programs to the CPP disability program. In addition, the increase in the denial rates could have increased the amount of self-screening done by applicants or reduced the potential for moral hazard on the part of applicants that report hard-to-diagnose medical conditions. These factors may account, to varying degrees, for our findings. One implication of these estimates is that it might be more appropriate to develop a menu of alternative estimates that correspond to different policy regimes rather than focusing efforts on producing a definitive estimate of the effect of denial rates on labor market outcomes.

Acknowledgements This research was supported by the Connaught Fund at the University of Toronto and the Social Sciences Humanities Research Council of Canada.

Published online: 9 January 2007

References

Agency for Health Care Policy and Research. (1994). Acute low back problems in adults. Rockville, MD.

Baker, M., & Fortin, N. (2001). Occupational gender composition and wages in Canada, 1987-1988. Canadian Journal of Economics. 34(2), 345 376 (May).

Bolduc, D., Fortin, B., Labrecque, F., & Lanoie, P. (2002). Workers' compensation, moral hazard and the composition of workplace injuries. Journal of Human Resources, 37(3), 623-652 (May).

Bound, J., & Burkhauser, R. V. (1999). Economic analysis of transfer programs targeted on people with disabilities. In O. Ashenfelter & D. Card (Eds.), Handbook of labor economics volume 3 (pp. 34173528). Amsterdam, The Netherlands: North Holland.

Campolieti, M. (2001). Disability insurance and the labour force participation of older men and women in Canada. Canadian Public Polio; 27(2), 179-194 (June).

Campolieti, M. (2002a). Moral hazard and disability insurance: On the incidence of hard-to-diagnose conditions in the Canada/Quebec pension plan disability program. Canadian Public Policy, 28(3), 419-441 (September).

Campolieti, M. (2002b). Disability insurance denial rates and the incidence of hard-to-diagnose medical conditions. Mimeograph. Centre for Industrial Relations, University of Toronto.

Campolieti, M. (2003). Disability insurance eligibility criteria and the labor supply of older men. Economics Bulletin, 10(3), 1-7 (July).

Campolieti, M. (2004). Disability insurance benefits and labor supply: Some additional evidence. Journal of Labor Economics. 22(4), 863-889 (November).

Campolieti, M., & Lavis, J. N. (2000). Disability expenditures in Canada, 1970-1996: Trends, reform efforts and a path for the future. Canadian Public Polio; 26(2), 241-264 (June).

Canada Pension Plan Consultations. (1996). An information paper for the consultations on the Canada pension plan. Ottawa, Ontario.

Donald, S. G., & Lang, K. (2001). Interference with difference in differences and other panel data. Mimeograph. Boston University, Department of Economics.

Frank, J., & Maetzel, A. (2000). Determining occupational disorder: Can this camel carry more straw? In T. Sullivan (Ed.), Injury and the new world of work (pp. 265-283). Vancouver, British Columbia: University of British Columbia Press.

Gruber, J., & Kubik, J. D. (1997). Disability insurance rejection rates and the labor supply of older workers. Journal of Public Economics. 64(1), 1-23 (April).

Halpern, J., & Hausman, J. A. (1986). Choice under uncertainty: A model of applications for the social security disability insurance program. Journal of Public Economics. 31(3), 131-161 (November).

Human Resources Development Canada. (1996). Evaluation of the Canada pension plan (Disability Component). Final Report. Ottawa, Ontario.

Kreider, B. (1999). Social security disability insurance: Applications, awards, and lifetime income flows. Journal of Labor Economics, 17(4), 784-827 (November).

Kreider, B., & Riphahn, R. T. (2000). Explaining applications to the U.S. disability insurance system: A semiparametric approach. Journal of Human Resources. 35(1), 82-115 (Winter).

Kushner, R. F. (1993). Body weight and mortality. Nutrition Reviews, 51(3), 127-136 (March).

Loeb, S., & Bound, J. (1996). The effect of measured school inputs on academic achievement: Evidence from the 1920s, 1930s and 1940s birth cohorts. Review of Economics and Statistics. 78(4), 653-664 (November).

Moulton, B. R. (1990). An illustration of a pitfall in estimating the effects of aggregate variables on micro units. Review of Economies and Statistics, 72(2), 334-338 (May).

Parsons, D. O. (1991). Self-screening in targeted public transfer programs. Journal of Political Economy, 99(4), 859-876 (August).

Quebec Task Force on Spinal Disorders. (1987). Scientific approach to the assessment and management of activity-related spinal disorders. Spine, 12(7S) (July Special Issue).

Ruser, J. W. (1998). Does workers' compensation encourage hard to diagnose injuries? Journal of Risk and Insurance, 65(1), 101-124 (March).

Torjman, S. (2002). The Canada pension plan disability benefit. Ottawa: Caledon Institute of Social Policy.

Vaillancourt, F. (2000). The Quebec pension plan: Institutional arrangements and lessons for Alberta. In P. Boothe (Ed.), A separate pension plan for Alberta: Analysis and discussion (pp. 23-40). Edmonton, Alberta: University of Alberta Press.

Wooldridge, J. (2003). Cluster-sample methods in applied econometrics. American Economic Review. 93 (2), 133-138 (May).

(1) Bound and Burkhauser (1999) provide a very thorough survey of this literature.

(2) Other papers examining the effects of denial rates on application behavior include Halpern and Hausman (1986), Kreider (1999) and Kreider and Riphahn (2000).

(3) Prior to September 1995 individuals aged 55-64 were considered disabled if they could not do the particular job they held rather than any job, while individuals younger than 55 were considered disabled if they could not do any job.

(4) The move to regional offices was part of HRDC's Income Security Program Redesign, which reorganized the delivery of the CPP as well as Old Age Security. These measures were intended to make these programs more responsive to the clientele as well as the local region (Torjman, 2002).

(5) The information on the denial rates for the CPP disability program was obtained from HRDC. The QPP denial rates for 1995-2000 were obtained from the Regie de Rentes du Quebec; prior to 1995 the QPP denial rates were computed using information in the annual reports of the Regie de Rentes du Quebec.

(6) HRDC aggregates the figures for Nova Scotia and Prince Edward Island as well as those for Manitoba and Saskatchewan; the other six provinces are presented individually.

(7) Vaillancourt (2000) presented some regression results that also suggest welfare beneficiaries were moving onto the CPP disability rolls during the early 1990s. Vaillancourt argued that tiffs shifting was driven by a change in the funding of provincial welfare programs, which were previously 50-50 cost shared with the federal government. This change made some provinces pay more than a 50 percent share if the growth in their expenditures exceeded a cap and thereby providing them with an incentive to shift beneficiaries onto another program.

(8) Source: unpublished data from HRDC. Figures for the other provinces and other years could not be obtained.

M. Campolieti

University of Toronto at Scarborough, Scarborough, Canada

M. Campolieti ([mail envelope]) * J. Goldenberg

Centre for Industrial Relations and Human Resources, University of Toronto,

121 St. George Street, Toronto, ON M5S 2E8, Canada

e-mail: campolie@chass.utoronto.ca

Table 1 C/QPP disability program denial rates

                                        Year

                                        1992   1993   1994   1995
1. CPP Program
   CPP Denial Rate (Program Level)      47     47     51     56
   CPP Denial Rate by Province/Region
   Newfoundland                         N/A    N/A    N/A    N/A
   Nova Scotia/Prince Edward Island     N/A    N/A    N/A    N/A
   New Brunswick                        N/A    N/A    N/A    N/A
   Ontario                              N/A    N/A    N/A    N/A
   Manitoba/Saskatchewan                N/A    N/A    N/A    N/A
   Alberta                              N/A    N/A    N/A    N/A
   British Columbia                     N/A    N/A    N/A    N/A
2. QPP Program
   QPP Denial Rate                      57     57     49     54

                                        Year

                                        1996   1997   1998   1999
1. CPP Program
   CPP Denial Rate (Program Level)      65     70     66     65
   CPP Denial Rate by Province/Region
   Newfoundland                         71     58     60     58
   Nova Scotia/Prince Edward Island     63     67     60     61
   New Brunswick                        59     55     57     61
   Ontario                              72     74     67     67
   Manitoba/Saskatchewan                71     68     66     60
   Alberta                              68     70     64     66
   British Columbia                     64     69     64     67
2. QPP Program
   QPP Denial Rate                      56     55     54     54

All numbers reported in the table are percentages. Denial Rate tot
CPP (Program Level) represents denial rate weighted by applications,
not the average of denial rates in each province/region. The data
for the denial rates at the province/region level are not available
prior to 1996.

Table 2 Summary statistics for individual characteristics, by gender

Variable Name               NPHS Data

                            Males             Females

                            Mean (Standard    Mean (Standard
                            Deviation)        Deviation)

Non-participation           0.2332 (0.4229)   0.4207 (0.4937)
Age 45-49                   0.3063 (0.4610)   0.2929 (0.4551)
Age 50-54                   0.2669 (0.4424)   0.2654 (0.4416)
Age 55-59                   0.2233 (0.4165)   0.2270 (0.4189)
Age 60-64                   0.2035 (0.4027)   0.2147 (0.4106)
Married                     0.7531 (0.4312)   0.6701 (0.4702)
No High School              0.2965 (0.4567)   0.2849 (0.4514)
Attended High School        0.1478 (0.3549)   0.1767 (0.3140)
Some Post Secondary         0.3521 (0.4777)   0.3683 (0.4824)
University                  0.2036 (0.4027)   0.1701 (0.3758)
BM1<22 or BMI>28            0.4057 (0.4910)   0.4625 (0.4986)
  (Disability Status/
  Poor Health)
Sample Size                 12,832            14,084
  (Number of Individuals)

Variable Name               SCF Data

                            Males             Females

                            Mean (Standard    Mean (Standard
                            Deviation)        Deviation)

Non-participation           0.2405 (0.4274)   0.4333 (0.4955)
Age 45 49                   0.3200 (0.4665)   0.3229 (0.4676)
Age 50 54                   0.2692 (0.4436)   0.2666 (0.4422)
Age 55 59                   0.2166 (0.4112)   0.2157 (0.4113)
Age 60-64                   0.1943 (0.3950)   0.1948 (0.3961)
Married                     0.8585 (0.3486)   0.7870 (0.4094)
No High School              0.3643 (0.4812)   0.3729 (0.4836)
Attended High School        0.1511 (0.3582)   0.1938 (0.3953)
Some Post Secondary         0.3347 (0.4719)   0.3351 (0.4720)
University                  0.1498 (0.3569)   0.0983 (0.2977)
BMI<22 or BMI>28            --                --
  (Disability Status/
  Poor Health)
Sample Size                 37,422            38,150
  (Number of Individuals)

Table 3 Estimates for the effect of DI denial rates on labor force
participation, second-step of Wooldridge's minimum distance estimator.
based on data from the NPHS

                       Including QPP Program
Males Aged 45--64
  Denial rate          -0.265 (0.118)   -0.119 (0.106)   -0.677 (0.189)
  Unemployment Rate    --               0.007  (0.002)   --
  Province ummies      No               No               Yes
  Year Dummies         No               No               Yes
  Number of Groups     18               18               18
  Overidentification   43.36 {0.997}    26.53 {0.967}    1.45 {0.016}
    Test Statistic
Females Aged 45-64
  Denial Rate          -0.424 (0.104)   -0.266 (0.095)   0.090 (0.427)
  Unemployment Rate    --               0.007 (0.002)    --
  Province dummies     No               No               Yes
  Year Dummies         No               No               Yes
  Number of Groups     18               18               18
  Overidentification   36.14 {0.997}    22.69 {0.909}    6.70 {0.540}
    Test Statistic

                       Including
                       QPP Program     Excluding QPP Program
Males Aged 45--64
  Denial rate          0.646 (0.166)   -0.149 (0.152)   0.006 (0.126)
  Unemployment Rate    0.002 (0.001)   --               0.007 (0.002)
  Province ummies      Yes             No               No
  Year Dummies         Yes             No               No
  Number of Groups     18              15               15
  Overidentification   0.95 {0.012}    36.72 {0.999}    19.16 {0.915}
    Test Statistic
Females Aged 45-64
  Denial Rate          0.020 (0.297)   -0.260 (0.196)   -0.192 (0.109)
  Unemployment Rate    0.006 (0.002)   --               0.007 (0.002)
  Province dummies     Yes             No               No
  Year Dummies         Yes             No               No
  Number of Groups     18              15               15
  Overidentification   3.30 {0.230}    28.83 {0.993}    15.86 {0.802}
    Test Statistic

                       Excluding QPP Program
Males Aged 45--64
  Denial rate          0.685 (0.193)   0.658 (0.255)
  Unemployment Rate    --              0.002 (0.008)
  Province ummies      Yes             Yes
  Year Dummies         Yes             Yes
  Number of Groups     15              15
  Overidentification   0.95 {0.081}    0.94 10.081}
    Test Statistic
Females Aged 45-64
  Denial Rate          0.200 (0.323)   0.250 (0.421)
  Unemployment Rate    --              0
  Province dummies     Yes             Yes
  Year Dummies         Yes             Yes
  Number of Groups     15              15
  Overidentification   2.89 {0.424}    2.26 {0.480}
    Test Statistic

The numbers presented in the table for the denial rate are the
coefficient estimates on the denial rate variable from the second
stage of Wooldridge's procedure. Standard errors are presented in
parentheses. The estimates in the columns including QPP program
include observations on the CPP provinces/adjudication regions as
well observations from Quebec. which is covered by the QPP program.
The estimates in the columns excluding the QPP program only include
observations on the CPP provinces/adjudication regions. The number
of groups indicates the number of adjudication regions. p values
for chi-squared distribution are presented in curly braces.

Table 4 Estimates for the effect of D1 denial rates on labor force
participation, second-step of Wooldridge's minimum distance estimator.
based on data from the SCF

                       Including QPP Program
Males Aged 45-64
  Denial Rate          -0.058 (0.107)   0.224 (0.059)    -0.524 (0.986)
  Unemployment Rate    --               0.014 (0.002)    --
  Province dummies     No               No               Yes
  Year Dummies         No               No               Yes
  Number of Groups     20               20               20
  Overidentification   240.80 {0.999}   47.32 {0.999}    88.82 {0.998}
    Test Statistic
Females Aged 45-64
  Denial Rate          -0.101 (0.122)   -1.203 (0.073)   -0.181 (0.322)
  Unemployment Rate    --               -1.017 (0.002)   --
  Province Dummies     No               No               Yes
  Year Dummies         No               No               Yes
  Number of Groups     20               20               20
  Overidentification   331.2 {0.999}    76.97 {0.999}    114.35 {0.999}
    Test Statistic

                       Including
                       QPP Program     Excluding QPP Program
Males Aged 45-64
  Denial Rate          0.122 (0.227)   -0.078 (0.129)   0.290 (0.054)
  Unemployment Rate    0.013 (0.001)   --               0.015 (0.001)
  Pro\ nice dummies    Yes             No               No
  Year Dummies         Yes             No               No
  Number of Groups     20              16               16
  Overidentification   3.89 {0.207}    243.06 {0.999}   25.17 {0.978}
    Test Statistic
Females Aged 45-64
  Denial Rate          0.213 (0.330)   -0.063 (0.128)   0.301 (0.050)
  Unemployment Rate    0.015 (0.002)   --               0.016 (0.001)
  Province Dummies     Yes             No               No
  Year Dummies         Yes             No               No
  Number of Groups     20              16               16
  Overidentification   9.22 {0.763}    269.18 {0.999}   24.30 {0.972}
    Test Statistic

                       Excluding QPP Program
Males Aged 45-64
  Denial Rate          -0.432 (1.062)   -0.179 (0.083)
  Unemployment Rate    --               0.006 (0.004)
  Pro\ nice dummies    Yes              Yes
  Year Dummies         Yes              Yes
  Number of Groups     16               16
  Overidentification   26.72 {0.994}    0.97 {0.085}
    Test Statistic
Females Aged 45-64
  Denial Rate          -0.013 (0.358)   -0.038 (0.276)
  Unemployment Rate    --               0.002 (0.012)
  Province Dummies     Yes              Yes
  Year Dummies         Yes              Yes
  Number of Groups     16               16
  Overidentification   31.48 {0.999}    10.34 {0.965}
    Test Statistic

The numbers presented in the table for the denial rate are the
coefficient estimates on the denial rate variable from the second
stage of Wooldridge's procedure. Standard errors are presented in
parentheses. The estimates in the columns including QPP program
include observations on the CPP provinces/adjudication regions as well
observations from Quebec, which is covered by the QPP program. The
estimates in the columns excluding the QPP program only include
observations on the CPP provinces adjudication regions. The number
of groups indicates the number of adjudication regions. p values
for chi-squared distribution are presented in curly braces.