06 December, 2016

Project Overview

Background and research question

Background

  • Current literature shows association with negative outcomes
    • Economic upticks and downticks associated with injury and unemployment rate
  • Potential relationship with disability duration
    • Extending the theoretical model, we hypothesize an association with disability outcomes

Research Question

What is the association between industry-level economic/labour market changes and disability days for working adults in the USA?

Methods

Data and approach

Data Sources: MEPS

Medical Expenditures Panel Survey (MEPS) used for disability and demographic data

MEPS survey characteristics

  • Annual survey of 15,000 households (civilian, non-institutionalized population)
  • Information on health care use, expenditures, demographics, access to care and health care quality
  • Surveys: Individual, Household, Medical provider, Insurance

MEPS stuctured as panel survey with longitudinal components

Example data collection timeline 2004 - 2006

Relevant variables available across years

Relatively large sample sizes indicate potential for meaningful level of power in analysis

YEAR obs
2003 34,215
2004 34,403
2005 33,961
2006 34,145
2007 30,964
2008 33,066
2009 36,855
2010 32,846
2011 35,313
2012 38,974
2013 36,940
2014 34,875

Relevant variables available across years

Relatively large sample sizes indicate potential for meaningful level of power in analysis

Var Description Records
DDNWRK Days Missed from Work 1,249,671
EMPST Employment Status 1,249,671
INDCAT Industry Category 1,249,671
MNHLTH Mental Health Status 1,249,671
REGION Region 1,249,671

Economic Data: Bureau of Labor Statistic (BLS)

Quarterly Census of Employment and Wages provides necessary economic variables

Economic Data: Bureau of Labor Statistic (BLS)

Quarterly Census of Employment and Wages provides necessary economic variables

Economic Data: Bureau of Labor Statistic (BLS)

Quarterly Census of Employment and Wages provides necessary economic variables

We can merge MEPS and BLS data on year and industry

Resulting dataset can be filtered and processed for analysis

Exclusion criteria: We only exclude observations which meet the following criteria

  • Employment status is not applicable (i.e. excluding students, retirees)
  • Days Sick from Work is a non-integer value
  • Industry category is undefined
  • Industry category is Government or Military (due to mapping and private/public distinction)

Data processing steps

  • Create adjusted variables from financial covariates
  • Create lag variables from labour market and financial covariates
  • Sum or average BLS covariates across sub-regions to create annual region-linked data (i.e. Northeast, Midwest, South, West)
  • Merge on year and industry for each observation of the MEPS dataset

Covariates have weak multilevel relationship

Individuals, household and potentially industries could have random effects

Dependent Variable: Days missed from work due to illness/disability

Covariates have weak multilevel relationship

Individuals, household and potentially industries could have random effects

Independent Variables Extracted/Available

  • Change in employment level from prior year (absolute, percentage)
  • Change in adjusted and non-adjusted average wage level from prior year (absolute, percentage)
  • Change in adjusted and non-adjusted total wages paid from prior year (absolute, percentage)
  • Industry
  • Gender
  • Race/Ethnicity
  • Average level of mental health
  • Region

Analysis will explore results from GLMs and LMMs

Models will be assessed by DIC, AIC, and other measures of fit

Packages to be used

  • glm for generalized mixed models
  • lmer for linear mixed models

Results

Analysis and Outcomes

Generalized Linear Models

Assess models with and without percentage change and interaction terms

Model 1: Absolute change, no interaction terms

\[SickDays ~ Industry + region + yr + lag(Emp) + lag(AvgWage) + lag(TotWage)\]

Model 2: Percentage change, no interaction terms

\[SickDays ~ Industry + region + yr + lag(Emp%) + lag(AvgWage%) + lag(TotWage%)\]

Model 3: Absolute change, year as interaction term

\[SickDays ~ Industry + region + yr*lag(Emp) + yr*lag(AvgWage) + yr*lag(TotWage)\]

Model 4: Percentage change, year as interaction term

\[SickDays ~ Industry + region + yr*lag(Emp%) + yr*lag(AvgWage%) + yr*\%lag(TotWage%)\]

MODEL 1:

Coefficients:
                                                Estimate Std. Error t value Pr(>|t|)    
(Intercept)                                    7.097e+01  1.581e+01   4.489 7.15e-06 ***
INDDESCEducation, Health, and Social Services  1.082e+00  1.320e-01   8.195 2.53e-16 ***
INDDESCFinancial Activities                    5.444e-01  1.545e-01   3.524 0.000425 ***
INDDESCInformation                             1.023e+00  2.236e-01   4.576 4.74e-06 ***
INDDESCLeisure and Hospitality                 6.306e-03  1.452e-01   0.043 0.965353    
INDDESCManufacturing                           7.083e-01  1.407e-01   5.035 4.78e-07 ***
INDDESCMining                                  1.770e-01  4.289e-01   0.413 0.679868    
INDDESCNatural Resources                       1.690e-01  2.353e-01   0.718 0.472707    
INDDESCOther Services                          2.140e-01  1.570e-01   1.363 0.172909    
INDDESCProfessional and Business Services      2.084e-02  1.567e-01   0.133 0.894174    
INDDESCTransportation and Utilities            1.875e+00  1.879e-01   9.977  < 2e-16 ***
INDDESCWholesale and Retail Trade              3.371e-01  1.588e-01   2.123 0.033786 *  
regionNE                                      -1.720e-02  9.583e-02  -0.179 0.857570    
regionS                                       -2.584e-01  8.277e-02  -3.122 0.001794 ** 
regionW                                       -4.197e-01  8.211e-02  -5.111 3.20e-07 ***
year                                          -3.404e-02  7.868e-03  -4.326 1.52e-05 ***
chg.employed                                  -2.020e-07  5.531e-07  -0.365 0.714927    
chg.avg.income.adj                             1.303e-05  4.949e-05   0.263 0.792363    
chg.total.wages.adj                            1.224e-11  1.015e-11   1.205 0.228069    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

AIC: 1457154

MODEL 2

Coefficients:
                                               Estimate Std. Error t value Pr(>|t|)    
(Intercept)                                   48.670648  17.708238   2.748 0.005988 ** 
INDDESCEducation, Health, and Social Services  1.150003   0.120266   9.562  < 2e-16 ***
INDDESCFinancial Activities                    0.575878   0.152022   3.788 0.000152 ***
INDDESCInformation                             1.018953   0.220426   4.623 3.79e-06 ***
INDDESCLeisure and Hospitality                -0.021575   0.133465  -0.162 0.871580    
INDDESCManufacturing                           0.773238   0.131028   5.901 3.61e-09 ***
INDDESCMining                                  0.056458   0.430102   0.131 0.895565    
INDDESCNatural Resources                       0.081550   0.235593   0.346 0.729230    
INDDESCOther Services                          0.176248   0.153946   1.145 0.252266    
INDDESCProfessional and Business Services      0.142712   0.130135   1.097 0.272796    
INDDESCTransportation and Utilities            1.982779   0.159348  12.443  < 2e-16 ***
INDDESCWholesale and Retail Trade              0.444552   0.124246   3.578 0.000346 ***
regionNE                                      -0.015260   0.095318  -0.160 0.872805    
regionS                                       -0.223449   0.078159  -2.859 0.004252 ** 
regionW                                       -0.422579   0.082600  -5.116 3.12e-07 ***
year                                          -0.022998   0.008797  -2.614 0.008945 ** 
pct.chg.employed                              -1.924867   2.959367  -0.650 0.515414    
pct.chg.avg.income.adj                         0.623679   3.919381   0.159 0.873569    
pct.chg.total.wages.adj                        2.895408   3.078582   0.941 0.346962    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

AIC: 1457151

MODEL 3

Coefficients:
                                                Estimate Std. Error t value Pr(>|t|)    
(Intercept)                                   -1.045e+02  3.612e+01  -2.892 0.003826 ** 
INDDESCEducation, Health, and Social Services  9.747e-01  1.337e-01   7.291 3.09e-13 ***
INDDESCFinancial Activities                    5.590e-01  1.547e-01   3.614 0.000301 ***
INDDESCInformation                             1.153e+00  2.246e-01   5.133 2.85e-07 ***
INDDESCLeisure and Hospitality                -5.296e-02  1.461e-01  -0.362 0.717024    
INDDESCManufacturing                           7.193e-01  1.442e-01   4.990 6.05e-07 ***
INDDESCMining                                  2.461e-01  4.294e-01   0.573 0.566636    
INDDESCNatural Resources                       2.468e-01  2.358e-01   1.047 0.295192    
INDDESCOther Services                          1.870e-01  1.573e-01   1.189 0.234570    
INDDESCProfessional and Business Services     -5.673e-02  1.581e-01  -0.359 0.719639    
INDDESCTransportation and Utilities            1.749e+00  1.912e-01   9.147  < 2e-16 ***
INDDESCWholesale and Retail Trade              2.167e-01  1.627e-01   1.332 0.182733    
regionNE                                      -2.092e-02  9.601e-02  -0.218 0.827523    
regionS                                       -3.179e-01  8.341e-02  -3.811 0.000138 ***
regionW                                       -4.261e-01  8.218e-02  -5.185 2.17e-07 ***
year                                           5.337e-02  1.799e-02   2.967 0.003006 ** 
chg.employed                                   5.004e-04  2.239e-04   2.235 0.025390 *  
chg.avg.income.adj                             5.945e-02  2.037e-02   2.918 0.003523 ** 
chg.total.wages.adj                            4.094e-09  3.432e-09   1.193 0.232995    
year:chg.employed                             -2.495e-07  1.114e-07  -2.240 0.025118 *  
year:chg.avg.income.adj                       -2.963e-05  1.014e-05  -2.922 0.003483 ** 
year:chg.total.wages.adj                      -2.025e-12  1.707e-12  -1.186 0.235711    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

AIC: 1457121

MODEL 4

Coefficients:
                                                Estimate Std. Error t value Pr(>|t|)    
(Intercept)                                   -2.294e+02  5.982e+01  -3.834 0.000126 ***
INDDESCEducation, Health, and Social Services  1.122e+00  1.207e-01   9.292  < 2e-16 ***
INDDESCFinancial Activities                    5.587e-01  1.522e-01   3.670 0.000243 ***
INDDESCInformation                             1.107e+00  2.226e-01   4.973 6.59e-07 ***
INDDESCLeisure and Hospitality                 1.783e-02  1.337e-01   0.133 0.893878    
INDDESCManufacturing                           7.849e-01  1.347e-01   5.825 5.71e-09 ***
INDDESCMining                                 -4.857e-02  4.323e-01  -0.112 0.910540    
INDDESCNatural Resources                       8.386e-02  2.372e-01   0.354 0.723683    
INDDESCOther Services                          2.325e-01  1.545e-01   1.505 0.132388    
INDDESCProfessional and Business Services      1.791e-01  1.310e-01   1.368 0.171342    
INDDESCTransportation and Utilities            2.026e+00  1.597e-01  12.687  < 2e-16 ***
INDDESCWholesale and Retail Trade              4.926e-01  1.247e-01   3.950 7.83e-05 ***
regionNE                                      -5.539e-02  9.569e-02  -0.579 0.562720    
regionS                                       -2.618e-01  7.856e-02  -3.333 0.000860 ***
regionW                                       -4.577e-01  8.274e-02  -5.532 3.17e-08 ***
year                                           1.154e-01  2.978e-02   3.876 0.000106 ***
pct.chg.employed                              -6.411e+03  1.822e+03  -3.519 0.000433 ***
pct.chg.avg.income.adj                        -3.114e+03  2.296e+03  -1.356 0.175023    
pct.chg.total.wages.adj                        8.077e+03  1.730e+03   4.667 3.05e-06 ***
year:pct.chg.employed                          3.188e+00  9.060e-01   3.518 0.000434 ***
year:pct.chg.avg.income.adj                    1.548e+00  1.143e+00   1.354 0.175607    
year:pct.chg.total.wages.adj                  -4.017e+00  8.609e-01  -4.666 3.07e-06 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

AIC: 1457096

MODEL 5: WITHOUT INDUSTRY

Coefficients:
                               Estimate Std. Error t value Pr(>|t|)    
(Intercept)                  -2.688e+02  5.934e+01  -4.531 5.88e-06 ***
regionNE                     -4.617e-02  9.552e-02  -0.483 0.628813    
regionS                      -2.978e-01  7.828e-02  -3.805 0.000142 ***
regionW                      -5.411e-01  8.246e-02  -6.562 5.34e-11 ***
year                          1.354e-01  2.954e-02   4.585 4.54e-06 ***
pct.chg.employed             -6.460e+03  1.773e+03  -3.644 0.000268 ***
pct.chg.avg.income.adj       -2.344e+03  2.282e+03  -1.027 0.304344    
pct.chg.total.wages.adj       8.260e+03  1.712e+03   4.824 1.41e-06 ***
year:pct.chg.employed         3.213e+00  8.816e-01   3.644 0.000269 ***
year:pct.chg.avg.income.adj   1.164e+00  1.136e+00   1.025 0.305500    
year:pct.chg.total.wages.adj -4.109e+00  8.520e-01  -4.823 1.42e-06 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for gaussian family taken to be 145.2018)

    Null deviance: 27095241  on 186464  degrees of freedom
Residual deviance: 27073458  on 186454  degrees of freedom
AIC: 1457424

Number of Fisher Scoring iterations: 2

LMMs

Assess models with and without percentage change and interaction terms and random effects

Models assessed

## With year as interaction term for economic variables
lmer.1 <- lmer(DDNWRKTOT ~ (1|DUPERSID) + INDCAT + region + year*pct.chg.total.wages.adj, data=df,
               REML = FALSE)
lmer.2 <- lmer(DDNWRKTOT ~ (1|DUPERSID) + region + year*pct.chg.total.wages.adj, data=df,
               REML = FALSE)
lmer.3 <- lmer(DDNWRKTOT ~ (1 |DUPERSID) + (1 | INDCAT) + region + year*pct.chg.total.wages.adj, data=df,
               REML = FALSE)
lmer.4 <- lmer(DDNWRKTOT ~ (1 |DUPERSID) + (1 | INDCAT) + (1 | region) + year*pct.chg.total.wages.adj, data=df,
               REML = FALSE)

## With year as an interaction term for industry too
lmer.5 <- lmer(DDNWRKTOT ~ (1 |DUPERSID) + (1 | INDCAT) + year*INDCAT + year*pct.chg.total.wages.adj, data=df,
               REML = FALSE)

## Without year as interaction term

kable(anova(lmer.1, lmer.2, lmer.3, lmer.4, lmer.5), format = 'markdown')
Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
lmer.4 8 1455016 1455097 -727499.8 1455000 NA NA NA
lmer.2 9 1455271 1455362 -727626.3 1455253 0.00000 1 1e+00
lmer.3 10 1455009 1455110 -727494.5 1454989 263.63606 1 0e+00
lmer.1 20 1454980 1455182 -727469.8 1454940 49.34197 10 4e-07
lmer.5 29 1455004 1455298 -727473.2 1454946 0.00000 9 1e+00

Discussion

Conclusions and limitations

  • Mixed model makes a difference, even in the case of what's generally a panel/cross-sectional dataset. The extent to which this effect is attributable to the 2-year
  • Including industry does improve AIC

The results of this study are limited by theory and data quality

Theory linking covariates and depth of region and industry data likely bias results

Economic indicator-related questions and limitations

  • Is percentage or absolute change in economic variables more appropriate?
  • How closely are sub-industries trends associated with grouped trends and can we generalize?
  • Is there a lag factor? If so, how long? How do we verify what is appropriate to use?

Next Steps

Variables to assess for inclusion

  • Gender
  • Ethnicity
  • Age
  • Level of Education

Methodological changes

  • Develop methods to asssess two year spurts for longitudinal panel participants