Background and Rationale

The overall goal of this study is to determine if having a disability (arthritis) can affect someone’s employment opportunity in the workplace. In 2011, an estimated 37.9 million people, 12.2% of the U.S. population, were living with a disability. Rheumatoid arthritis is estimated to be present in 1.3 million U.S. adults 18 years or older, representing 0.6% of the population based on NHIS-and NHANES-derived analyses from the National Arthritis Data Workgroup (Y, Chan, & Carruthers 2014). The relationship of this consists of a random sample comparing those who have arthritis to those who do not and the causal effect of each individual’s employment. Those with a condition such as arthritis may struggle with social isolation and conflict with others. By examining these measures, it can help improve functioning and help prevent unemployment for those with disabilities. It is important to address this question because of the discrimination that takes place in the work environment. Individuals with a disability are often discriminated against especially when there is a preconceived conception on their condition. They are typically viewed as being less capable when it comes to knowledge, skills, and the ability to perform the job at hand. Persons with disabilities may be less productive and incur higher insurance costs for the company as well.

Study Design

The data used in this analysis came from the 2017 Behavioral Risk Factor Surveillance System and Association of People Supporting Employment First (“Map of Employment First Across the nation”). The BRFFS contains a wide variety of health-related surveys in the United States. This study uses cross-sectional variation of Employment First to examine systematic differences in the employment outcome of individuals associated with a disability or not. I am specifically targeting males and females with arthritis, ages 39-59. For this particular study, employ1_cl is the key outcome variable and is measured by whether an individual is employed or not employed. This variable is binary meaning those who are employed are represented by a 1 and those who are not by a 0. The map represents the key independent variable, Policy, and measures which states have some implementation of the Employment First policy or not. There are currently 19 states that have the Employment First policy and it is continuing to grow nationally. The original map for the policy was separated into six different categories: No known activity, Activity no policy, Directive, Executive order, Legislation, and Legislation plus Directive or Executive order. I took the three variables Executive order, Legislation, and Legislation plus Directive or Executive order and classified this as “has the policy.” The remaining three were then classified as “does not have the policy.” This variable then became binary and was coded by (1/0).

Emplpoyment First Policy by State

Setting

The Employment First policy has been implemented among various states across the United States beginning in the early 20th century and leading to the present day. The policy has been in the works for many decades and continues to grow at an average rate. States are begging to adopt the policy in hopes it improves the employment rate for the disabled and brings more opportunities to the overall working community. The U.S Department of Labor’s Office of Disability Employment Policy (ODEP) created the Employment First State Leadership Mentor Program to help align policies, regulations, and funding to encourage integrated employment for individuals with significant disabilities (Company News 2015). The state of Washington has concluded that the Employment First policy does make a difference. For example, in the area of employment services for individuals with intellectual and developmental disabilities, Washington State has had a long-standing commitment to policies and practices focused on employment in the community as the first priority. The end result is that 89% of individuals served by the Washington State system are in integrated employment services compared to a national average of 20%. Many states are even well below this average with some at less than 10% (APSE Fact Sheet).

Participants and Variables

I have chosen to include the following individuals from the BRFSS in my analysis to help provide the best estimate for the outcome variable. For my study, the target population are those both male and female specifically ages 39-59. Arthritis typically starts at the age of 40 and before 60 years of age, so focusing on this particular age range is adequate. I will be studying those who have arthritis and their geographical region they subsidies in. My outcome variable (employ1_cl) is whether an individual is employed or not employed. The Employment First policy is coded by X.STATE and whether it has some aspect of the policy or not (1/0). male_cl specifies whether the individual is male or female, ARTHDIS2 states whether arthritis affects the individual’s work or not, MSCODE provides the geographical region of the individual and helps control bias, and lastly, X.AGEG5YR helps narrow down the population to those who are 39-59 years old.

#Filter the BRFSS data to include only [fill in criteria]
#This runds on hard drive only
#load("brfss17.RData")
#analysis <- brfss17 %>% filter(X.AGEG5YR >= 5, X.AGEG5YR <= 8) %>% select(X.STATE, ARTHDIS2, SEX, X.AGEG5YR, MSCODE, EMPLOY1)
#save(analysis, file = "analysis.RData")

#Select BRFSS Variables
load("analysis.RData")
 
 

#Save analysis Data to take to Cloud


#Recode Outcome and take out people who do not have arthritis
analysis <- analysis %>% filter(ARTHDIS2 != 7 & ARTHDIS2 != 9 & !is.na(ARTHDIS2)) %>% mutate(arthis2_cl = ifelse(ARTHDIS2 == 1|ARTHDIS2 == 2, ARTHDIS2, NA))
analysis <- analysis %>% mutate(arthis2_cl = ifelse(arthis2_cl == 2, 0, arthis2_cl))


#Recode additional variables
analysis <- analysis %>% filter(SEX != 9) %>% mutate(male_cl = ifelse(SEX == 1, 1, 0))
analysis <- analysis %>% mutate(mscode_cl = ifelse(MSCODE == 1| MSCODE == 2| MSCODE == 3| MSCODE == 5, MSCODE, "NA"))

analysis <- analysis %>% mutate(x.ageg5yr_cl = ifelse(X.AGEG5YR == 5| X.AGEG5YR == 6| X.AGEG5YR == 7| X.AGEG5YR == 8, X.AGEG5YR, "NA" ))

analysis$x.ageg5yr_cl <- as.numeric(analysis$x.ageg5yr_cl)
analysis <- analysis %>% mutate(x.ageg5yr_cl = x.ageg5yr_cl - 4) 

analysis <- analysis %>% mutate(employ1_cl = ifelse(EMPLOY1 == 1|EMPLOY1 == 2, 1, 0))


#Join your policy data with the BRFSS Data
PolicyData <- read.csv("PolicyData.csv")
analysis <- full_join(analysis, PolicyData, by = "X.STATE")

Study Size

There are 144,278 observations in my analysis data set and 15 variables. There was a total of 309,327 observations eliminated due to restricting the sample to only persons with arthritis and missing data from the BRFSS.

Statistical Methods

To examine the relationship if whether the Employment First policy increases the number of individuals employed, I compare the employment of those with arthritis which is provided by the BRFSS data and if the state has some implementation of the policy. I begin with model 1 which is a simple regression.

\[employ1_cl_{i} = \beta_{0} + \beta_{1}policy_{i}+\epsilon_{i}\]

In model 1, the policy variable is represented by \(\beta_{1}\), for x represents those individuals who do or do not live in a state with the policy. This is denoted as 1 for people living in a state with the Employment First policy and 0 for people living in states without the policy. The binary variable employ1_cl is identified as the Y variable. A positive estimate for B1x1, meaning the state has the policy, would imply a positive correlation of the increased number of individuals being employed with arthritis. This method attributes differences in employment between states with varying Employment First policy. This may produce a biased estimate if the states differ in employment for other reasons.

Model 2 \[employ1_cl_{i} = \beta_{0} + \beta_{1}policy_{i}+ \beta_{2}X.AGEG5YR + \beta_{3}male_cl + \epsilon_{i}\]

To help control bias in the estimates, I included age in my multiple regression. Age can negatively correlate to arthritis because most individuals with arthritis tend to be older. This may play a biased role in employment for individuals specifically 39-59 years of age. I also included sex which is represented by the variable male_cl. Dependent on whether an individual is male or female can lead to a biased estimate of the effect of the policy on the employment outcome. Females are typically discriminated against in comparison to males and are also known to suffer from arthritis more (verywellhealth 2018).

Model 3

\[employ1_cl_{i} = \beta_{0} + \beta_{1}policy_{i}+ \beta_{2}X.AGEG5YR + \beta_{3}male_cl+\beta_{4}policy_{i}*male_cl + \beta_{5}MSCODE + \epsilon_{i}\]

In this multiple regression, I have multiplied policy by male_cl to create an interaction between the two variables. This shows whether the policy has a different relationship with males. Also, I have included the variable MSCODE which represents the geographical region of where an individual lives. Dependent on a person’s region, they may not have been reachable when BRFSS was conducting the survey therefore hindering the results. Region can also lead to bias when specifically looking at which states have implemented the policy as well. Does the geographical region have a higher population? Was the policy implemented in richer states? These questions are important to ask and take into consideration when estimating their potential relationship leading to a biased estimate.

Descriptive Statistics

#Include a table of descriptive statistics for your outcome variable and x variables. If your policy is discrete you can include separate columns for each value of your policy variable.

# analysis %>% summarize(avg_employ1 = mean(employ1_cl, na.rm = TRUE))
# analysis %>% group_by() 
# analysis %>% ggplot(aes(x = ARTHDIS2)) + geom_histogram(SEX = 1)

analysis$Policy <- as.factor(analysis$Policy)
plotdata <- analysis %>% group_by(Policy) %>% summarize(employment.rate = mean(employ1_cl, na.rm = TRUE))
plotdata %>% filter(!is.na(Policy)) %>% ggplot(aes(x = Policy, y = employment.rate)) + geom_col()

Table 1

In this table I am comparing the employment rate to the policy variable (employ1_cl). The policy is coded as 0 (does not have the policy) and 1 (does have the policy). One can see that the states who do not have the policy and those who do, the mean is relatively the same at .312 or 31%. This shows that even with states that have some implementation of the policy, it does not greatly impact the employment rate.

Results

#Include a table of regressions using stargazer to create a nice summary of your output

simple <- lm(employ1_cl ~ Policy, data = analysis)
  
multiple1 <- lm(employ1_cl ~ Policy + as.factor(X.AGEG5YR) + male_cl, data = analysis)
  
multiple2 <- lm(employ1_cl ~ Policy + as.factor(X.AGEG5YR) + male_cl + Policy*male_cl, data = analysis)

multiple3 <- lm(employ1_cl ~ Policy + as.factor(X.AGEG5YR) + male_cl + Policy *male_cl + as.factor(MSCODE), data = analysis)

#multiple4 <- lm(arthis2_cl ~ Policy + as.factor(X.AGEG5YR) + male_cl + Policy *male_cl, data = analysis)

#stargazer table for use while writing
stargazer(simple, multiple1, multiple3, header = FALSE, type = "text")

======================================================================================================== Dependent variable:
——————————————————————————— employ1_cl
(1) (2) (3)
——————————————————————————————————– Policy1 0.0002 0.001 -0.001
(0.003) (0.002) (0.003)

as.factor(X.AGEG5YR)2 0.084*** 0.041
(0.018) (0.053)

as.factor(X.AGEG5YR)3 0.060*** 0.092*
(0.017) (0.048)

as.factor(X.AGEG5YR)4 0.053*** 0.088*
(0.016) (0.046)

as.factor(X.AGEG5YR)5 0.052*** 0.100**
(0.015) (0.045)

as.factor(X.AGEG5YR)6 0.040*** 0.122***
(0.015) (0.044)

as.factor(X.AGEG5YR)7 0.003 0.078*
(0.014) (0.044)

as.factor(X.AGEG5YR)8 -0.041*** 0.025
(0.014) (0.044)

as.factor(X.AGEG5YR)9 -0.151*** -0.067
(0.014) (0.044)

as.factor(X.AGEG5YR)10 -0.335*** -0.264***
(0.014) (0.044)

as.factor(X.AGEG5YR)11 -0.419*** -0.344***
(0.014) (0.044)

as.factor(X.AGEG5YR)12 -0.469*** -0.387***
(0.014) (0.044)

as.factor(X.AGEG5YR)13 -0.514*** -0.429***
(0.014) (0.044)

as.factor(X.AGEG5YR)14 -0.230*** -0.209***
(0.017) (0.046)

male_cl 0.077*** 0.054***
(0.002) (0.004)

as.factor(MSCODE)2 0.014***
(0.004)

as.factor(MSCODE)3 0.008**
(0.004)

as.factor(MSCODE)5 0.001
(0.003)

Policy1:male_cl 0.010*
(0.006)

Constant 0.313*** 0.535*** 0.449***
(0.002) (0.014) (0.044)

Observations 141,843 141,843 77,495
R2 0.00000 0.203 0.194
Adjusted R2 -0.00001 0.203 0.194
Residual Std. Error 0.464 (df = 141841) 0.414 (df = 141827) 0.383 (df = 77475)
F Statistic 0.006 (df = 1; 141841) 2,411.162*** (df = 15; 141827) 980.222*** (df = 19; 77475) ======================================================================================================== Note: p<0.1; p<0.05; p<0.01

#stargazer(simple)
#stargazer table for knit version - comment out the one above and comment this one in
stargazer(simple, multiple1, multiple3, header = FALSE, title = "Table 2", type = "html")

Table 2

When looking at the overall table of the three regressions, the coefficients show that the policy has no statistical significance of improving the employment rate with those states that have the Employment First policy. When specifically looking at the policy variable across columns, the coefficients are close to zero which concludes that there is no statistical significance in improving the employment rate. What stands out is column three when controlling for the variable Policy1:male_cl. This has the largest coefficient stating that those states with the Employment First policy, increases employment by 1 percentage points for males or 3.33% making it statistically significant. Even with a 3.33% increase this may not be practically significant.

Robustness Checks

surveyweights <- lm(employ1_cl ~ Policy + as.factor(X.AGEG5YR) + male_cl + Policy *male_cl + as.factor(MSCODE), data = analysis, weights = X.LLCPWT)
#   
# hetrobust <-
# 
# #stargazer table for use while writing
# stargazer(multiple2, surveyweights, hetrobust, header = FALSE, type = "text")
# 
# #stargazer table for knit version - comment out the one above and comment this one in
stargazer(multiple3, surveyweights, header = FALSE, title = "Table 3", type = "text")

Overview of Key Findings

This study was conducted to examine the potential impact of those who have a disability (arthritis) and whether they were employed or not in the workplace. With the policy Employment First, this allowed those with disabilities an opportunity to work receiving comparable wages and work without being discriminated against. Prior research has shown that those with arthritis struggle in finding work and maintaining a job, but with the Employment First policy, it could help reduce this negative connotation. I hypothesized that this policy would have an increase in employment rates among states who have implemented the Employment First policy, but evidence shows that there is no statistical significance that this policy increases employment rates among those with arthritis. This analysis does have some potential limitations. First, the BRFFS data was conducted using a phone survey. Although I have filtered down my data to a more specific population, I still had enough observations to conduct my study. Potential factors hindering the results are those who represent the older population (39-59 years old) and those who may be of lower income who did not have a landline. Another factor is geographical region and the individuals access to a landline in that region. The MSCODE represents regions throughout the country and is dependent on an individual’s area code. If they lived in a particular area that had cellphone access only, they were unreachable when conducting the survey immediately eliminating from the data being collected. Policy makers should help Employment First continue to grow and make sure all individuals suffering from a disability are aware of this policy. I believe that over time, Employment First could help benefit those with a disability and become a national policy helping increase the overall employment rate among states.