Does Unemployment Affect the Percentage of People that Voted for Donald Trump in 2016?
By Ainsley Cox
The Cooperative Election Study is a stratified sample survey administered by YouGov and Harvard University. The survey collects over 50,000 responses to common content and specific content questions two times every election year, with a pre-election wave and a post-election wave. Data from this survey in 2016 found that out of all of the people who responded that they were unemployed on that survey, 45.8% responded that they were Democrats and 25% responded that they were Republicans. This made me wonder, is there a relationship between ones political party and their employment status?
In order to answer this question, I wanted to take a look at two of the most polarizing elections years in recent history to try and make sure that there would be a lower chance of people voting for the other parties candidate. I took data from the election years 2016 and 2020 for the percentage of people who voted for Donald Trump and the percentage of unemployment in those areas.
In my analysis, I merged the dataset on unemployment with the US Counties dataset on the state columns. In order to do this, I had to do some manual data cleaning in Excel. Originally, I had wanted to join both of these datasets via the county and state columns. However, in the unemployment dataset, values looked like this — ‘Bibb County, AL’ while in the US counties dataset, they looked like this — ‘Bibb, Alabama’. I attempted to find a way to join on these columns, however I was unable to. As an alternative, I decided to join on state rather than on state and county.
After merging the dataset, I realized that the Unemployment dataset when tested with the number of people who voted for Trump in 2016, was not statistically significant. However, I did see that there was a column with the percent of unemployment in the data set with the percentage of people who voted for Trump in 2016. From here, I decided to conduct my tests using these.
I created two lists with 4 historically Democratic and Republican states. The Democratic states were California, New York, Colorado, Massachusetts, and Vermont. The Republican states were Texas, Florida, Wyoming, Georgia, and Alabama. Once I had made these lists, I used a Pearson R test to determine the correlation coefficient and the P-value of the relationship among these states between unemployment and the number of people that voted for Trump in 2016.
The visualizations the resulted from this analysis aren’t the best and cannot be understood that intuitively, but they do communicate some interesting findings.
As you can see, the results are a lot more spread out within the Republican counties than in the Democratic counties. It is commonly known that amongst most states, there are more Republican leaning counties with much smaller populations than the few Democratic counties, so it would make sense that there are more points on the Republican states graph. Overall, from these graphs, there are more Republican states whose counties voted highly for Donald Trump, but there is an even scatter of unemployment within both of these visualizations.
The correlation coefficient for Democratic states was -0.035962416784547035, meaning that the relationship between Unemployment and voting for Trump in 2016 was a negligible negative relationship and not statistically significant. However, the correlation coefficient for Republican states was -0.2691811752420403, meaning that there is a strong negative relationship between the percent of unemployment in a county and the percentage of people who voted For Donald Trump in 2016 in Republican states. This means that in these states, as the percentage of unemployed people goes down, the percentage of people that voted for Donald Trump goes up.
After discovering this, I decided to run a regression on all of the values and not just those specific states.
From the graph, it seems like a lesser negative correlation, but a negative correlation none the less. The correlation coefficient for all states between the percentage of people who voted for Donald Trump in 2016 and the percentage of people who are unemployed, it came out to -0.20401942706555287. This means that this is again a significant negative correlation. So on average, as the percentage of unemployed people decreases, the percentage of people who voted for Donald Trump in 2016 increases.
There could be various confounding variables that this analysis isn’t taking into account, for example the number of people per county widely varies and the number of people per state varies even more. There are more conservative counties than liberal counties, but the liberal counties tend to have much higher populations than the conservative ones. The same can be said for US states.
However, it is more likely that this is in fact a causal relationship and not a random one. Older people tend to hold more conservative values than younger people, and older people tend to have higher levels of employment than younger people do. Therefore, it would make sense that as the level of unemployment decreases in a state or county, the average age of someone in that county would increase, thus the county would become more conservative and there would be more votes cast for conservative candidates like Donald Trump.
According the The Cooperative Election Study, more Democrats are unemployed because there are more Democrats in the U.S. than any other political party. From this Data analysis, we see that as the rate of unemployment decreases, voters tend to vote for more conservative candidates.
