Saturday, January 13, 2018

Illness and Employment After Becoming Eligible for Social Security Benefits

Illness and Employment After Becoming Eligible for Social Security Benefits

Situation:  Under current law, people can start receiving Social Security retirement benefits at age 62. However, people who claim benefits at age 62 receive 75 percent of the full retirement benefit.  The full benefit is not available until age 66. The actual benefit as a share of the full benefit increases by around 0.4 or 0.5 percentage points for each month a person delays claiming benefits.  Spousal benefits are also reduced for people who claim early.

Go here for a description of Social Security benefits.


Questions: People with illnesses often must leave work prior to becoming eligible for the full Social Security benefit.  

How does illness impact the ability of a person to remain in the workforce long enough to obtain the full Social Security benefit? 

Does illness directly impact work status or does illness alter the impact of additional years worked past age 62 on work status?


Data:   The source of data for this investigation is the medical expenditures panel survey for 2015.  Here is the codebook.


The population studied involves all people between the age of 62 and 66.   62 is the first year a person can obtain the Social Security Retirement benefit.  66 is the year people currently get the full benefit.

The source for the dependent variable is variable EMPST31.   A dummy variable was set to 1 for people not employed during the survey round and 0 otherwise.

The explanatory variables in the model include information on diseases inflicting individuals, education, and time elapsed after reaching age 62.

The model included two educational variables – a no-college dummy variable and a college degree dummy variable.   The omitted class is some college.  It is expected that education should allow the person to stay employed past age 62.

The disease dummy variables included in this analysis are – diabetes, stroke, asthma, angina, arthritis, cancer, and emphysema.

The age variable was constructed with year and month of birth date compared to midyear 2015

Methodology:

Three versions of logit regression models where the dependent variables is the not employed dummy were considered and compared.   Not employed includes people who are retired and people who may be looking for a position but do not have one.

The first version includes dummy variables on education and health and information on age over 62.

The second version includes previous variables and interactions between the age over 62 and disease variables.  

The third version excludes the direct disease variables but allows disease to impact the age coefficient through interaction variables.

For each model, I present the odds ratio on the coefficient and the p-value for the test the odds ratio is different from 1.0.

An odds ratio greater than one indicates a variable increases the probability a person is not employed while an odds ratio less than one indicates the variable decreases the probability a person is unemployed.

Results:

Results for the three models are presented below.


Three Models Where the Dependent
 Variable is Not Employed in Round
Model One:  Direct impacts Only
Exlanatory Variable
Odd Ratio
p-value
diabetic
1.762
0
stroke
2.370
0
asthma
1.293
0.158
angina
1.444
0.147
arthritis
1.594
0
cancer
1.237
0.126
emphysema
2.331
0.006
college
0.648
0.084
no_college
1.298
0.045
addyear
1.249
0
_cons
0.363
0
Model Two:  Direct Impacts and Interaction Variables
Exlanatory Variable
Odd Ratio
p-value
diabetic
1.757
0
stroke
2.595
0.004
asthma
1.624
0.121
angina
0.819
0.664
arthritis
1.679
0.004
cancer
1.693
0.026
emphysema
1.897
0.224
college
0.662
0.1
no_college
1.282
0.058
addyear
1.302
0
stroke_addyear
0.949
0.707
asthma_addyear
0.887
0.341
angina_addyear
1.339
0.142
arthritis_addyear
0.970
0.684
cancer_addyear
0.852
0.096
emphysema_addyear
1.121
0.611
_cons
0.339
0
Model Three No Direct Disease Effects Interaction Variables Only
Exlanatory Variable
Odd Ratio
p-value
college
0.626
0.058
no_college
1.323
0.028
no_disease_addyear
0.988
0.792
stroke_addyear
1.387
0
asthma_addyear
1.077
0.312
angina_addyear
1.308
0.019
arthritis_addyear
1.208
0
cancer_addyear
1.043
0.445
emphysema_addyear
1.432
0.009
_cons
0.689
0.003


Analysis of results:

First Model:

The odds ratios for all seven disease variables are greater than 1.0.   Four of the p-values are significant at conventional levels.   There is a strong relationship between diseases and being out of the work force for people between the age of 62 and 66 inclusive.

The p-value for the college degree variable is significantly less than 1 while the p-value of the no-college variable is greater than 1.0.   These results indicate that education increases the odds of remaining in the workforce.  (Recall the base group or this analysis is people with some college but no degree.)

The odds ratio for the addyear variable is significantly greater than 1 indicating that increases in age over 62 increase the probability of a person not having a job.

Second Model:

The second model includes both direct disease effects and interactions between the  disease and  addyear variable.   Models with multiple dummy variables, which interact with a continuous explanatory variable are often plagued by collinearity.    The collinearity among the variables reduces the precision of coefficient estimates.

Six of the seven disease odds ratios are positive.  The asthma and emphysema p-values are no longer significantly different from 1.0.

In this model, the addyear odds ratio greater than 1.0 represents the basic increase in time past age 62 on whether a person is not working.   This odds ratio is in fact significantly greater than 1.0 indicating that passage of time after age 62 increases the likelihood a person does not have a job.

A disease/addyear interaction variable odds ratio greater than one suggests that the disease increases impact of aging on the decision to leave the workforce.   None of the odds ratios for the interaction variables are significantly greater than 1.0.  (There is a collinearity problem.)

Third Model:

The third model includes a no-disease/addyear interaction variable and separate disease/interaction variables.   

The no-disease/addyear odds ratio is NOT significantly different from 1.0.  (THIS IS  A REALLY INTERESTING RESULT.   PEOPLE WHO HAVE NO DISEASES ARE MUCH MORE LIKELY TO STAY IN THE WORKFORCE AFTER AGE 62.)

All the point estimates for the seven disease/addyear interaction variables are greater than 1.0.   Four of the seven odds ratios are significantly greater than 1.0.

Additional Econometric Work:   I am conducting additional work on this topic.   Work in progress includes statistical tests comparing alternative models, evaluation of impact of disease and age on employment status for different age groups, and stability of relationships over time and over the business cycle.

Potential Policy Implications:  Retirement experts suggest that many people must work more years to obtain a financially secure retirement.   This advice is hard to take for people leaving the work force because of their health status.

Policy analysts and politicians concerned about future Social Security shortfalls desire to create incentives and rules which will keep people in the workforce.  These changes would create significant hardships for people leaving the workforce due to their health status.