Donald Trump in Alabama

Donald Trump in Alabama

Roy Moore Speaking in Alabama

Roy Moore Speaking in Alabama

1 Introduction

Our research asks the question, “How did Alabama voter support for Donald Trump in the 2016 presidential election in the Black Belt vs. Non Black Belt regions correlate with and inform voter support in those regions for Roy Moore in the 2017 Senate election?” Our data comes from research cited in the Atlantic article, “African American Voters Made Doug Jones a US Senator in Alabama”.1 The 2017 Senate special election, to replace the seat vacated by Attorney General Jeff Sessions, was heavily covered by the media after the sexual assault allegations against front runner Roy Moore.2

The Black Belt is a historically black region of the South and will serve as a proxy variable for race. We used the map from the Center for Business and Economic Research at the University of Alabama to determine the traditional counties of the Black Belt.3 Historically, African Americans have shown greater support for Democratic candidates than Republican candidates.

Despite the fact that Trump won his election and Moore did not, we expect that region (Black Belt vs. Non Black Belt) and support for Trump will inform support for Moore.

A limitation of our data set is that it does not fully represent the Alabama population’s voting preferences because it only includes those who actually voted. Furthermore, region classification (Black Belt vs. Non Black Belt) infers race, but is an imperfect proxy with potential inaccuracies.

Black Belt Counties in Alabama

Black Belt Counties in Alabama

2 Exploratory data analysis

First six rows of our data set:

region county candidate percent_votes_received
NotBB Autauga Trump 0.7277
NotBB Autauga Moore 0.5990
NotBB Baldwin Trump 0.7655
NotBB Baldwin Moore 0.6170
BB Barbour Trump 0.5210
BB Barbour Moore 0.4200

Each row in this represents a different county and its voting percentage for either Trump or Moore, as well as whether it is in the Black Belt or Non Black Belt geographic region.

2.1 Summary Statistics

Summary of Percent Votes Trump Received
region mean std_dev
BB 0.3877294 0.1646162
NotBB 0.7251460 0.1115340
Summary of Percent Votes Moore Received
region mean std_dev
BB 0.3118235 0.1471607
NotBB 0.6203600 0.1279218

Trump received a greater share of votes in both regions compared to Moore. The similar standard deviations show comparable spreads across regions and elections. The decrease in vote percentage for the Republican candidates from 2016 to 2017 in both regions shows declining party support. However, Moore and Trump did both receive a much greater share of votes in the Non Black Belt region compared to the Black Belt region, as seen in the following two histograms.

2.2 Graphs

Each point on the scatterplot represents a county in Alabama. The graph shows voter support for Trump on the x-axis and voter support for Moore on the y-axis. The color of the points reflect whether it is a county in the Black Belt or Non Black Belt region. There is a positive correlation between voting for Trump and voting for Moore. Counties that had high support for Trump tended to have high support for Moore and vice versa for counties with low support. This graph does not effectively show the impact of region on voting support because the black and blue points are intermixed.

This boxplot shows that both Moore and Trump’s median share of votes across Non Black Belt counties were considerably higher than in Black Belt counties. Trump received greater median support than Moore within each region. The graph also shows that the spread of votes for both candidates was greater in Black Belt counties.

3 Multiple regression

term estimate std_error statistic p_value conf_low conf_high
intercept -3.259 1.745 -1.867 0.067 -6.747 0.229
Trump 0.888 0.042 21.341 0.000 0.805 0.971
regionNotBB -15.617 3.111 -5.020 0.000 -21.833 -9.401
Trump:regionNotBB 0.228 0.054 4.179 0.000 0.119 0.336

3.1 Statistical interpretation

We used an interaction model for the multiple regression because there are different slopes for each region.

Predicted BB % Votes Moore = -3.259 + 0.888 (BB % Votes Trump)
Predicted NonBB % Votes Moore = -18.876 + 1.116 (NonBB % Votes Trump)

The slopes represent the associated effect of Percent Votes for Trump on Percent Votes for Moore depending on the level of region (Black Belt vs. Non Black Belt). Our model shows that in the Black Belt region, a one percentage point increase in votes for Trump has, on average, an associated increase of 0.888 of a percentage point votes for Moore. For the Non Black Belt region, our model shows that a one percentage point increase in votes for Trump has, on average, an associated increase of 1.116 percentage points in votes for Moore. The negative intercepts in both models show that, in a hypothetical sense, if Trump received no votes, Moore would receive a negative percent of the votes. Negative votes are impossible, so this phenomenon instead shows that Moore received a lower percentage of votes on average than did Trump in both regions.

3.2 Non-statistical interpretation

Our model shows that counties in which Trump received a greater share of votes often corresponded with Moore also receiving a greater share of votes. Both candidates tended to do better in the Non Black Belt counties.

4 Inference for multiple regression

As displayed in the regression table in Section 3, the confidence interval for the slope relating percent votes Moore received based on those Trump received in the Black Belt region is [0.805, 0.971]. We can conclude with 95% certainty that the true slope value falls within that interval. Because 0 does not fall in this interval, we are confident that there is a positive relationship between Trump and Moore votes in the Black Belt region. Similarly, the confidence for the slope relating percent votes Moore received based on those Trump received in the Non Black Belt region is [0.924, 1.307]. We can similarly conclude with 95% certainty that the true slope value falls within that interval. This interval has a greater upper limit than that for the Black Belt region, so it is possible that there is a steeper correlation in the Non Black Belt.

The null hypotheses would be that there is no relationship between votes received by Trump and votes received by Moore. Since the p-values are near zero, we can reject the null hypotheses. Therefore, our predictors are statistically significant. They do show a relationship between the two variables.

In our initial scatterplot, we observed a linear relationship between voter support for Trump and voter support for Moore. This satisfies the first requirement for inference. The histogram of residuals shows that they are fairly normal. The Black Belt region’s residuals are slightly left skewed, but this is partly due to its being such a small sample size with only n=17 counties. The Non Black Belt counties’ residuals are more normal, in part because of the larger size of n=50 counties. This satisfies the second requirement for inference.

The scatterplot of residuals shows somewhat even spread. About an even number of points, in both facets, appear on either side of the line y=0. In the Black Belt region, there are a few more points on the positive side of the line than the negative side of the line, but as before, that can be due in part to the smaller n=17 sample size. In the Non Black Belt region, spread appears more constant and symmetrical about the blue line, y=0. This satisfies the third and final requirement for inference. Thus, we can draw appropriate conclusions from our regression model, but keeping in mind that the residuals are not perfectly normal or spread.

5 Conclusion

In our analysis, we tried to determine how voter support for Donald Trump in the 2016 presidential election correlated with and informed voter support for Roy Moore in the 2017 Senate election in the Black Belt and Non Black Belt regions of Alabama. Our analysis indicates there was a positive correlation between votes for Trump and votes for Moore in both regions, meaning the greater share of votes Trump received the more likely Moore was to also receive a greater share of votes. This relationship was stronger in the Non Black Belt region, for the slope was steeper. We used the Black Belt region category as a proxy race, because the Black Belt tends to have a higher population of Black Americans. Our data shows that the Black Belt region showed less support for both Republican candidates studied compared to the rest of Alabama, indicating that voter race may have played a factors in the elections.

A few limitations of our analysis is that the voters in the presidential and senate elections were not necessarily the same individuals; therefore, we are comparing two distinct sample sets to a certain degree. Another limitation is that we used region classification (Black Belt vs. Non Black Belt) as a proxy for race, but we did not actually analyze the racial breakdown of the voter percentages. Thus, our analysis better illuminates voter support between historically, racially-divided regions rather than individuals. Furthermore, we did not account for county population size and only looked at percentages within each county. Therefore, we gave each county the same value in our analysis, even if one had significantly fewer or more citizens voting. Additionally, media coverage of Roy Moore’s sexual assault allegations may have played a role in his loss but we did not include this in our analysis.

In the future, we can analyze the actual racial breakdown of each county and study how different racial groups voted. This would allow us to draw more specific conclusions about the impact of race in the elections. This data could be compared with past and future election data in Alabama to see if this is a trend. If it is, it could indicate a systemic discrepancy between political views across race in Alabama.

6 Citations and References

6.1 Works Cited

  1. Newkirk, Van R. II. “African American Voters Made Doug Jones a U.S. Senator in Alabama.” Politics. The Atlantic, December 12, 2017. https://www.theatlantic.com/politics/archive/2017/12/despite-the-obstacles-black-voters-make-a-statement-in-alabama/548237/.

  2. McCrummen, Stephanie, Beth Reinhard, and Alice Crites. “Woman says Roy Moore initiated sexual encounter when she was 14, he was 32.” Investigations. The Washington Post, November 9, 2017. https://www.washingtonpost.com/investigations/woman-says-roy-moore-initiated-sexual-encounter-when-she-was-14-he-was-32/2017/11/09/1f495878-c293-11e7-afe9-4f60b5a6c4a0_story.html?utm_term.

  3. “Alabama Maps.” Center for Business and Economic Research at The University of Alabama. Accessed March 28, 2018. https://cber.cba.ua.edu/edata/maps/AlabamaMaps1.html.

7 Supplementary Materials

7.1 Watch Roy Moore ride his horse to the polls!

IMAGE ALT TEXT HERE