29 November, 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
| 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
glmfor generalized mixed modelslmerfor 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
GEE (Autoregressive)
Conduct GEE with and without interaction term with year
gee.1 <- geeglm(DDNWRKTOT ~ INDDESC + region + pct.chg.employed + pct.chg.avg.income.adj,
data = df, id = DUPERSID,
family = gaussian, corstr = "ar1")
gee.2 <- geeglm(DDNWRKTOT ~ INDDESC + region + year*pct.chg.employed + year*pct.chg.avg.income.adj,
data = df, id = DUPERSID,
family = gaussian, corstr = "ar1")
gee.3 <- geeglm(DDNWRKTOT ~ year*INDDESC + region + year*pct.chg.employed + year*pct.chg.avg.income.adj,
data = df, id = DUPERSID,
family = gaussian, corstr = "ar1")
summary(gee.1)
## ## Call: ## geeglm(formula = DDNWRKTOT ~ INDDESC + region + pct.chg.employed + ## pct.chg.avg.income.adj, family = gaussian, data = df, id = DUPERSID, ## corstr = "ar1") ## ## Coefficients: ## Estimate Std.err Wald ## (Intercept) 2.383297 0.140266 288.702 ## INDDESCEducation, Health, and Social Services 1.168795 0.121468 92.587 ## INDDESCFinancial Activities 0.562565 0.149507 14.159 ## INDDESCInformation 1.010286 0.252970 15.950 ## INDDESCLeisure and Hospitality -0.028944 0.126645 0.052 ## INDDESCManufacturing 0.764100 0.141798 29.038 ## INDDESCMining 0.118101 0.393617 0.090 ## INDDESCNatural Resources 0.087259 0.255908 0.116 ## INDDESCOther Services 0.185014 0.155402 1.417 ## INDDESCProfessional and Business Services 0.135840 0.123672 1.206 ## INDDESCTransportation and Utilities 1.970453 0.220948 79.533 ## INDDESCWholesale and Retail Trade 0.444239 0.125892 12.452 ## regionNE -0.008236 0.104658 0.006 ## regionS -0.205780 0.082274 6.256 ## regionW -0.412180 0.086809 22.544 ## pct.chg.employed 0.019717 0.978024 0.000 ## pct.chg.avg.income.adj 5.445677 1.350403 16.262 ## Pr(>|W|) ## (Intercept) < 2e-16 *** ## INDDESCEducation, Health, and Social Services < 2e-16 *** ## INDDESCFinancial Activities 0.000168 *** ## INDDESCInformation 6.51e-05 *** ## INDDESCLeisure and Hospitality 0.819225 ## INDDESCManufacturing 7.10e-08 *** ## INDDESCMining 0.764147 ## INDDESCNatural Resources 0.733120 ## INDDESCOther Services 0.233829 ## INDDESCProfessional and Business Services 0.272035 ## INDDESCTransportation and Utilities < 2e-16 *** ## INDDESCWholesale and Retail Trade 0.000418 *** ## regionNE 0.937272 ## regionS 0.012379 * ## regionW 2.05e-06 *** ## pct.chg.employed 0.983916 ## pct.chg.avg.income.adj 5.52e-05 *** ## --- ## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 ## ## Estimated Scale Parameters: ## Estimate Std.err ## (Intercept) 145 3.824 ## ## Correlation: Structure = ar1 Link = identity ## ## Estimated Correlation Parameters: ## Estimate Std.err ## alpha 0.1595 0.009298 ## Number of clusters: 104677 Maximum cluster size: 5
summary(gee.2)
## ## Call: ## geeglm(formula = DDNWRKTOT ~ INDDESC + region + year * pct.chg.employed + ## year * pct.chg.avg.income.adj, family = gaussian, data = df, ## id = DUPERSID, corstr = "ar1") ## ## Coefficients: ## Estimate Std.err Wald ## (Intercept) -261.2439 54.5102 22.97 ## INDDESCEducation, Health, and Social Services 1.0917 0.1219 80.18 ## INDDESCFinancial Activities 0.5374 0.1493 12.96 ## INDDESCInformation 1.1339 0.2550 19.77 ## INDDESCLeisure and Hospitality -0.0415 0.1266 0.11 ## INDDESCManufacturing 0.8030 0.1448 30.77 ## INDDESCMining 0.0810 0.3945 0.04 ## INDDESCNatural Resources 0.0959 0.2558 0.14 ## INDDESCOther Services 0.1921 0.1557 1.52 ## INDDESCProfessional and Business Services 0.1349 0.1239 1.19 ## INDDESCTransportation and Utilities 1.9646 0.2209 79.07 ## INDDESCWholesale and Retail Trade 0.4387 0.1261 12.11 ## regionNE -0.0401 0.1048 0.15 ## regionS -0.2516 0.0825 9.31 ## regionW -0.4443 0.0869 26.16 ## year 0.1313 0.0271 23.43 ## pct.chg.employed 1723.4749 556.3211 9.60 ## pct.chg.avg.income.adj 6029.3128 1044.7010 33.31 ## year:pct.chg.employed -0.8577 0.2768 9.60 ## year:pct.chg.avg.income.adj -3.0008 0.5201 33.29 ## Pr(>|W|) ## (Intercept) 1.6e-06 *** ## INDDESCEducation, Health, and Social Services < 2e-16 *** ## INDDESCFinancial Activities 0.00032 *** ## INDDESCInformation 8.7e-06 *** ## INDDESCLeisure and Hospitality 0.74291 ## INDDESCManufacturing 2.9e-08 *** ## INDDESCMining 0.83736 ## INDDESCNatural Resources 0.70769 ## INDDESCOther Services 0.21711 ## INDDESCProfessional and Business Services 0.27621 ## INDDESCTransportation and Utilities < 2e-16 *** ## INDDESCWholesale and Retail Trade 0.00050 *** ## regionNE 0.70205 ## regionS 0.00228 ** ## regionW 3.1e-07 *** ## year 1.3e-06 *** ## pct.chg.employed 0.00195 ** ## pct.chg.avg.income.adj 7.9e-09 *** ## year:pct.chg.employed 0.00195 ** ## year:pct.chg.avg.income.adj 8.0e-09 *** ## --- ## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 ## ## Estimated Scale Parameters: ## Estimate Std.err ## (Intercept) 145 3.82 ## ## Correlation: Structure = ar1 Link = identity ## ## Estimated Correlation Parameters: ## Estimate Std.err ## alpha 0.16 0.00929 ## Number of clusters: 104677 Maximum cluster size: 5
summary(gee.3)
## ## Call: ## geeglm(formula = DDNWRKTOT ~ year * INDDESC + region + year * ## pct.chg.employed + year * pct.chg.avg.income.adj, family = gaussian, ## data = df, id = DUPERSID, corstr = "ar1") ## ## Coefficients: ## Estimate Std.err ## (Intercept) -272.1621 74.6570 ## year 0.1367 0.0372 ## INDDESCEducation, Health, and Social Services -78.5200 63.4022 ## INDDESCFinancial Activities -183.5714 83.3354 ## INDDESCInformation 31.7450 142.2602 ## INDDESCLeisure and Hospitality 41.2213 67.1469 ## INDDESCManufacturing 68.8779 83.1420 ## INDDESCMining -173.4727 202.8841 ## INDDESCNatural Resources -194.3351 140.8576 ## INDDESCOther Services 95.7924 85.1496 ## INDDESCProfessional and Business Services -135.1923 66.5431 ## INDDESCTransportation and Utilities 34.3324 116.7006 ## INDDESCWholesale and Retail Trade -25.5019 67.4557 ## regionNE -0.0453 0.1048 ## regionS -0.2634 0.0825 ## regionW -0.4564 0.0867 ## pct.chg.employed 2445.8987 672.0699 ## pct.chg.avg.income.adj 6708.6080 1098.6825 ## year:INDDESCEducation, Health, and Social Services 0.0396 0.0316 ## year:INDDESCFinancial Activities 0.0917 0.0415 ## year:INDDESCInformation -0.0152 0.0708 ## year:INDDESCLeisure and Hospitality -0.0205 0.0334 ## year:INDDESCManufacturing -0.0339 0.0414 ## year:INDDESCMining 0.0864 0.1011 ## year:INDDESCNatural Resources 0.0968 0.0702 ## year:INDDESCOther Services -0.0476 0.0424 ## year:INDDESCProfessional and Business Services 0.0674 0.0331 ## year:INDDESCTransportation and Utilities -0.0161 0.0581 ## year:INDDESCWholesale and Retail Trade 0.0129 0.0336 ## year:pct.chg.employed -1.2172 0.3344 ## year:pct.chg.avg.income.adj -3.3390 0.5470 ## Wald Pr(>|W|) ## (Intercept) 13.29 0.00027 *** ## year 13.54 0.00023 *** ## INDDESCEducation, Health, and Social Services 1.53 0.21555 ## INDDESCFinancial Activities 4.85 0.02761 * ## INDDESCInformation 0.05 0.82342 ## INDDESCLeisure and Hospitality 0.38 0.53928 ## INDDESCManufacturing 0.69 0.40742 ## INDDESCMining 0.73 0.39253 ## INDDESCNatural Resources 1.90 0.16769 ## INDDESCOther Services 1.27 0.26059 ## INDDESCProfessional and Business Services 4.13 0.04219 * ## INDDESCTransportation and Utilities 0.09 0.76861 ## INDDESCWholesale and Retail Trade 0.14 0.70539 ## regionNE 0.19 0.66571 ## regionS 10.20 0.00141 ** ## regionW 27.71 1.4e-07 *** ## pct.chg.employed 13.24 0.00027 *** ## pct.chg.avg.income.adj 37.28 1.0e-09 *** ## year:INDDESCEducation, Health, and Social Services 1.58 0.20930 ## year:INDDESCFinancial Activities 4.88 0.02716 * ## year:INDDESCInformation 0.05 0.82981 ## year:INDDESCLeisure and Hospitality 0.38 0.53894 ## year:INDDESCManufacturing 0.67 0.41293 ## year:INDDESCMining 0.73 0.39237 ## year:INDDESCNatural Resources 1.90 0.16763 ## year:INDDESCOther Services 1.26 0.26141 ## year:INDDESCProfessional and Business Services 4.14 0.04198 * ## year:INDDESCTransportation and Utilities 0.08 0.78155 ## year:INDDESCWholesale and Retail Trade 0.15 0.70043 ## year:pct.chg.employed 13.25 0.00027 *** ## year:pct.chg.avg.income.adj 37.26 1.0e-09 *** ## --- ## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 ## ## Estimated Scale Parameters: ## Estimate Std.err ## (Intercept) 145 3.82 ## ## Correlation: Structure = ar1 Link = identity ## ## Estimated Correlation Parameters: ## Estimate Std.err ## alpha 0.16 0.00929 ## Number of clusters: 104677 Maximum cluster size: 5
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 | -727500 | 1455000 | NA | NA | NA |
| lmer.2 | 9 | 1455271 | 1455362 | -727626 | 1455253 | 0.0 | 1 | 1 |
| lmer.3 | 10 | 1455009 | 1455110 | -727494 | 1454989 | 263.6 | 1 | 0 |
| lmer.1 | 20 | 1454980 | 1455182 | -727470 | 1454940 | 49.3 | 10 | 0 |
| lmer.5 | 29 | 1455004 | 1455298 | -727473 | 1454946 | 0.0 | 9 | 1 |
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