DATA 606 Project
Part 1 - Introduction
Is there is correlation between hate crimes during the election and the percentage of minority groups in each state? Racism is one of the biggest social issues all over the world. In the US, the rise of extremist groups and hate crimes have seen an increase since the election of our current president. With the controversy of the current election it is often suggested that hate crimes would occur in states that voted in favor of a controversial candidate such as (Donald Trump) with groups that voted in favor and those that voted against. The purpose of this project is to investivate a relationship between hate crimes during the period of November 9-18 2016, and the presence of minorities groups and presidential elect chosen per state across the US.
Part 2 - Data
Here is the datset we will be working with for this project.
## state share_non_white ElectTrump hate_crimes_per_100k_splc
## 1 Alabama 0.35 1 0.12583893
## 2 Alaska 0.42 1 0.14374012
## 3 Arizona 0.49 1 0.22531995
## 4 Arkansas 0.26 1 0.06906077
## 5 Georgia 0.48 1 0.12042027
## 6 Idaho 0.16 1 0.12420817
## 7 Indiana 0.20 1 0.24700888
## 8 Iowa 0.15 1 0.45442742
## 9 Kansas 0.25 1 0.10515247
## 10 Kentucky 0.15 1 0.32439697
## 11 Louisiana 0.42 1 0.10973335
## 12 Mississippi 0.44 1 0.06744680
## 13 Missouri 0.20 1 0.18452351
## 14 Montana 0.10 1 0.49549103
## 15 Nebraska 0.21 1 0.15948963
## 16 North Carolina 0.38 1 0.24400659
## 17 Ohio 0.21 1 0.19071396
## 18 Oklahoma 0.35 1 0.13362910
## 19 South Carolina 0.36 1 0.20989442
## 20 Tennessee 0.27 1 0.19993848
## 21 Texas 0.56 1 0.21358394
## 22 West Virginia 0.07 1 0.32867707
## 23 California 0.61 0 0.25580536
## 24 Colorado 0.31 0 0.39052330
## 25 Connecticut 0.30 0 0.33539227
## 26 Delaware 0.37 0 0.32275417
## 27 Florida 0.46 0 0.18752122
## 28 Illinois 0.37 0 0.19534455
## 29 Maine 0.09 0 0.61557402
## 30 Maryland 0.50 0 0.37043897
## 31 Massachusetts 0.27 0 0.63081059
## 32 Michigan 0.24 0 0.40377937
## 33 Minnesota 0.18 0 0.62747993
## 34 Nevada 0.50 0 0.14167316
## 35 New Hampshire 0.09 0 0.15154960
## 36 New Jersey 0.44 0 0.07830591
## 37 New Mexico 0.62 0 0.29481132
## 38 New York 0.42 0 0.35062045
## 39 Oregon 0.26 0 0.83284961
## 40 Pennsylvania 0.24 0 0.28510109
## 41 Rhode Island 0.28 0 0.09540164
## 42 Utah 0.19 0 0.13654673
## 43 Vermont 0.06 0 0.32414911
## 44 Virginia 0.38 0 0.36324890
## 45 Washington 0.31 0 0.67748765
## 46 Wisconsin 0.22 0 0.22619711Data collection: The data was collected through the github repository at https://raw.githubusercontent.com/fivethirtyeight/data/master/hate-crimes/hate_crimes.csv. It was transformed to be reduced to the table presented above. Also, the variable ElectTrump was included to show which states voted from Donald Trump.
Cases: The cases are 46 states, 4 of the 50 were removed due to reporting NA values for hate_crimes_per_100k_spplc.
Variables: The variables used for this experiment are share_non_white, ElectTrump, and hate_crimes_per_100k_spplc. share_non_white is the proportion of non-white voters per state (numerical). ElectTrump is whether the state voted for Donald Trump in the 2016 election (categorical). hate_crimes_per_100k_spplc is the hate crimes per 100k population per the Southern Poverty Law Center, from November 9-18 2016 (numerical).
Type of study: This is an observational study. We are not performing an experimental design, we are merely taking statistics that were reported and trying to find if there is any correlation between hate crimes, and minority population and the candidate elected in each state in the U.S.
Scope of inference - generalizability: The population of interest in this experiment is the United States. Any findings from this experiment can be generalized to the population since we have a random and large enough sample. However, I do believe the one factor that might be cause a source of bias is the fact that not the entire country voted. There is a large amount of citizens that did not vote that could’ve and should’ve voted which could also effect the variable share_non_white and even the ElectTrump variable.
Scope of inference - causality: This is possible, and I would argue that if hate groups support a candidate and that candidate wins the state, there is a possibility that hate crimes could rise. It would be very interesting to find a correlation between the 3 variables and might suggest that hate groups are active in the states, however that would be bringing in another variable into this project.
Part 3 - Exploratory data analysis
From this we can see that there is a right skew in the variable hate_crimes_per_100k_splc. We also have some outliers, and the data seems to be multimodal.
This is a subset of the original dataset where we’re only looking at states that voted for Donald Trump. We can see there is still a right skew in the data.
This is a subset of the original dataset looking at only states that did not vote for Donald Trump. It does not appear to be normally distributed either, however it does not appear to be skewed as much as the original or the previous subset.
There really isn’t any apparent trend, for the most part hate_crimes_per_100k_splc is under 0.4 as share_non_white increase. However, it is important to note that when hate_crimes_per_100k_splc is above 0.4, share_non_white is roughly under 0.3.
Looking at the 4 charts above, red represents Republican and blue represents Democrat. The states that voted republican seem to have a lower hate_crimes_per_100k_splc than those that voted democrat. There are only 2 republican states that had a hate_crimes_per_100k_splc over 0.4 while 5 democrats had it over 0.4. Also, 0 republican states had hate_crimes_per_100k_splc over 0.5, with only 1 (Missouri) approaching 0.5. Democrat states however had 5 with hate_crimes_per_100k_splc over 0.6! For the most part, republican states seem to be at or under 0.2 hate_crimes_per_100k_splc while democratic states seem to be at or under 0.4 hate_crimes_per_100k_splc. It also appears that democratic states tend to high a slightly hire share of non white voters on average.
#Build the Model
model <- lm(hate_crimes_per_100k_splc ~ share_non_white + ElectTrump, dataset)
summary(model)##
## Call:
## lm(formula = hate_crimes_per_100k_splc ~ share_non_white + ElectTrump,
## data = dataset)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.28156 -0.10193 -0.01254 0.06523 0.46410
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.46719 0.06123 7.630 1.6e-09 ***
## share_non_white -0.37862 0.16253 -2.330 0.02459 *
## ElectTrump1 -0.15218 0.04647 -3.275 0.00209 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.1567 on 43 degrees of freedom
## Multiple R-squared: 0.2568, Adjusted R-squared: 0.2222
## F-statistic: 7.429 on 2 and 43 DF, p-value: 0.001694We see from this model that both share_non_white and ElectTrump are significant predictors. While holding all other variables constant, we expect the estimation of hate_crimes_per_100k_splc to decrease by 0.37862 for 100% share_non_white voters. Also, while holding all other variables constant, we expect states that voted Donald Trump to estimate 0.15218 lower in hate_crimes_per_100k_splc than those who didn’t.
Part 4 - Inference
Linear Relationship
## Warning in abline(model): only using the first two of 3 regression
## coefficients
From this plot it appears we have a linear relationship betweent the independent variable and the dependent variable.
As mentioned earlier, it appears that democratic states have a higher mean in hate_crimes_per_100k_splc than republic states.
Normality
It appears that the residuals nearly follow normality in the model. There are some outliers present however.
We can see that the outliers have some influence at the tail end.
Multicollinearity
res <- cor(dataset[,c(2,4)])
round(res,2)## share_non_white hate_crimes_per_100k_splc
## share_non_white 1.00 -0.27
## hate_crimes_per_100k_splc -0.27 1.00It doesn’t appear that we have any multicollinearity in the variable hate_crimes_per_100k_splc.
Homoscedasticity
There appears to be constant variablity, and there is no heteroscedasticity.
Part 5 - Conclusion
The model appears to follow the conditions necessary. From it, it appears that while holding all other predictors constant, as the share of non white voters rises there is a decrease in hate crimes per 100,000 population by 0.37862. Also, while holding all other predictors constant, there is a decrease in hate crimes per 100,000 population by 0.15218. Ideally, the state with a higher population of non white voters as well as one that voted for Donald Trump estimated lower hate crimes per 100,000 population. Personally, I believed that these results would’ve been the opposite. I assumed the states that saw the most hate crimes voted Republican and had a low number of non white voters. However, these findings shed light on a new perspective. States that voted Democrat seem to be the ones with the most hate crimes even though they tend to have a higher share of non white voters on average. For further research I would suggest looking into whether these states have active hate groups, the amount of each type of hate crime that happened during November 9-18 2016, and also rerunning this again if the current president decides to rerun in 2020.