CovidRaceData

The CovidRaceData package contains two data sets on the COVID-19 pandemic in the US by race/ethnicity (as of 10/07/2020). More specifically, it has data for the number of COVID-19 cases and deaths by race/ethnicity for each state.

The raw data set (covid_race_data_raw) was obtained from The COVID Tracking Project and contains the number of COVID-19 cases and deaths by race/ethnicity by US state (as well as each state’s population).

The aggregated_covid_race_df contains information on the percent of cases/deaths by race/ethnicity vs. the percent of the population that each race/ethnicity composes in the US.

Installation

You can install the released version of CovidRaceData from GitHub with:

# install.packages("remotes")
remotes::install_github("Oliver-BE/CovidRaceData")

Example Usage

Visualization

Some examples of how to visualize the included data sets can be shown in the graph of COVID-19 cases per 100,000 people by race/ethnicity below:

library(CovidRaceData)
library(ggplot2)
library(reshape2)

# COVID-19 cases per 100,000 people by race or ethnicity
ggplot(aggregated_covid_race_df, aes(x = reorder(Race, Cases_Per_100K), y = Cases_Per_100K)) +
  geom_bar(stat = "identity", fill = "steelblue") +
  geom_text(aes(label = round(Cases_Per_100K, 0)),
          position = position_dodge(width=0.9), vjust=-0.25) +
  labs(x = "Race or ethnicity", y = "Cases per 100,000 people",
      title = "COVID-19 cases per 100,000 people by race or ethnicity")

Analysis

We can also use the provided data sets to answer questions such as whether or not there is a significant difference in the number of COVID-19 deaths by race or ethnicity in the United States. To do so, we can compare the distribution of the US population by race/ethnicity to the distribution of COVID-19 deaths by race/ethnicity. We begin with a visual representation:

# percentage of total COVID-19 deaths compared to percentage of total US population by race or ethnicity
df_deaths_population_by_percent <- melt(data.frame(aggregated_covid_race_df$Race,
                            aggregated_covid_race_df$Deaths_Percent,
                            aggregated_covid_race_df$Population_Percent)) 
colnames(df_deaths_population_by_percent) <- c("Race", "Variable", "Value")

ggplot(df_deaths_population_by_percent, aes(x = Race, y = Value, fill = Variable)) +
  geom_bar(stat = "identity", position = "dodge") +
  geom_text(aes(label = round(Value, 3)),
            position = position_dodge(width = 0.9), vjust = -0.25, size = 2.85) +
  scale_fill_discrete(labels = c("COVID-19 deaths", "Population")) +
  labs(x = "Race or ethnicity", y = "Percentage", fill = "Legend",
       title = "Percentage of total COVID-19 deaths compared to percentage of total US \n
       population by race or ethnicity")

Chi-Squared Test

To test if the distribution of the US population by race/ethnicity is significantly different than the distribution of COVID-19 deaths by race/ethnicity, we can carry out a chi-squared test. To carry out this test we first check our assumptions. We assume here that each data point is independent of one another and we can verify that 80% of expected counts are greater than 5. Thus, our assumptions are met.

Our null hypothesis here is that the distribution of our sample data (the number of COVID-19 deaths by race) matches the distribution of the population of the US by race.

Our alternative hypothesis here is that the distribution of our sample data does not match the distribution of the population of the US by race.

observed_frequency_deaths <- aggregated_covid_race_df$Deaths_Total
expected_frequency <- aggregated_covid_race_df$Population_Percent

chisq.test(observed_frequency_deaths, p = expected_frequency) 
#> 
#>  Chi-squared test for given probabilities
#> 
#> data:  observed_frequency_deaths
#> X-squared = 20960, df = 6, p-value < 2.2e-16

Our observed test statistic X^2 is 20960 and our p-value is 2.2e-16. Thus at a significance level of alpha = 0.01 we can reject the null hypothesis and conclude that there’s sufficient evidence to suggest that the distribution of the number of COVID-19 deaths by race does not match the distribution of the population of the US by race. This means that the distribution of COVID-19 deaths by race do not match what we would expect them to be (based on the distribution of population by race).

ANOVA Test

We can also carry out an ANOVA test to check for the difference in group means (which here is the number of deaths by race/ethnicity per 100 thousand people).

From the visualization above, we can see that there is a clear difference in the number of deaths per 100 thousand people across different races/ethnicities, so we proceed with our ANOVA test to test this statistically.

Our null hypothesis here is that all of the group means are equal (there’s an equal number of COVID-19 deaths per 100 thousand people for each race/ethnicity). Our alternative hypothesis is that the group means are not equal for all of the groups.

deaths_model <- lm(Deaths_Per_100K ~ Race, data = aggregated_covid_race_df)
anova(deaths_model)
#> Analysis of Variance Table
#> 
#> Response: Deaths_Per_100K
#>           Df Sum Sq Mean Sq F value Pr(>F)
#> Race       6 5008.2  834.69               
#> Residuals  0    0.0

We observe an essentially perfect fit for this ANOVA F-test, meaning that we can reject the null hypothesis and conclude that the number of COVID-19 deaths per 100 thousand people is significantly different across different races/ethnicities in the US. This matches what we saw in our visualization above, where there was very clearly a difference in the number of deaths between races (for example, 92 deaths for every 100 thousand Black people vs 51 for every 100 thousand white people).

Oliver-BE / CovidRaceData