Hypothesis Testing with Two Groups
Chapter 10: Hypothesis Testing with Two Groups
10.1 Getting Ready
This chapter focuses on hypothesis testing for comparing means and proportions between population subgroups.
Data Set: Load the
anes20.rdadata set in R.Required R Libraries: Ensure the following libraries are attached:
DescToolsHmiscgplotsdescreffectsize
10.2 Testing Hypotheses about Two Means
Hypothesis Testing Goal: Compare differences between two sample means rather than testing hypotheses about the overall population support.
Example: Exploring support differences across demographics like sex, race, and location impacts.
Gender Gap Analysis: Focus on how men and women differ in political attitudes, illustrated with the feeling thermometer rating for feminists.
Feeling Thermometer Rating: A scale from 0 (negative feelings) to 100 (positive feelings).
Expectations: Women are expected to rate feminists higher due to the intrinsic connection of feminism to women's rights.
10.2.1 Generating Subgroup Means
Use the
aggregatefunction to examine mean levels of support:Function Format:
aggregate(dependent, by=list(independent), FUN=stat_you_want).Implementation for Gender Gap:
r anes20$Rsex <- factor(anes20$V201600) levels(anes20$Rsex) <- c("Male", "Female") anes20$femFT <- anes20$V202160 agg_femFT <- aggregate(anes20$femFT, by=list(anes20$Rsex), FUN=(mean), na.rm=TRUE)The resulting output shows:
Male mean rating: 54.54
Female mean rating: 62.55
Conclusion: Women view feminists more positively than men by approximately 8 points.
10.3 Hypothesis Testing with Two Means
Null Hypothesis (H0): There is no relationship; population group means are equal: H_0: ext{μ}_1 = ext{μ}_2.
Alternative Hypotheses (H1):
Two-tailed: H_1: ext{μ}_1
eq ext{μ}_2.One-tailed (negative): H_1: ext{μ}_1 < ext{μ}_2.
One-tailed (positive): H_1: ext{μ}_1 > ext{μ}_2.
The task is to determine if the observed subgroup means differ significantly, not merely due to random variation.
10.3.1 A Theoretical Example
Logic of Hypothesis Testing: If H0 is true (no difference), we assess the likelihood of obtaining a sample difference
Transforming Differences: Convert sample difference into a t-score. For a sample:
t-score formula:
t = \frac{\bar{x}1 - \bar{x}_2 - (\mu_1 - \mu_2)}{S{\bar{x}_1 - \bar{x}_2}}
As the sample size increases, the t-distribution resembles the z-distribution, with critical values transitioning from 1.645 for one-tailed tests to 1.96 for two-tailed tests based on degrees of freedom.
10.3.2 Empirical Example Application
Applying the earlier findings on the feminist feeling thermometer for gender difference:
Null Hypothesis: H_0: \mu_W = \mu_M
Expected Direction: H_1: \mu_W > \mu_M.
t-score Calculation:
t-score formula adapted:
t = \frac{\bar{x}W - \bar{x}_M}{S{\bar{x}_1 - \bar{x}_2}}For example calculations using R:
r t = -12.89 # example t-score calculated
Critical Value Comparison:
If the absolute t exceeds the critical value (-1.645), reject the null hypothesis.
R Command for T-Test:
t.test(anes20$femFT ~ anes20$Rsex)produces:t = -13, df = 7111, p-value < 2e-16.
10.3.3 Calculating Effect Size
Importance of assessing practical significance vs statistical significance; introduce Cohen’s D as a measure of effect size:
D = \frac{\bar{x}1 - \bar{x}_2}{S{pooled}}
Comparison yields:
Feminist Feeling Thermometer: D = -0.2993
Big Business: D = 0.0494.
Effect Size Interpretation:
A small effect size for feminist feelings indicates limited practical implications in comparison to larger differences seen in that dimension vs. attitudes towards big business.
10.4 Difference in Proportions
Methodology for examining proportions between two groups:
Exploratory example on abortion attitudes, particularly respondents viewing abortion as illegal:
Null Hypothesis: H_0: \text{proportion of males}
eq \text{proportion of females}.
Sample Statistics: Proportion of males (0.1041) vs females (0.1074) suggests trivial differences; perform hypothesis testing on these proportions with a test similar to two means comparing.
Findings: Hypothesis test indicates no significant difference leading to a failure to reject H0.
10.5 Plotting Mean Differences
Comparisons can be visually represented through various plots:
Use Boxplots for distributions and Barplots for mean comparisons.
R Syntax for Plots:
barplot()andplotmeans()to visualize data clearly and aptly, labeling x/y axes accordingly.
Implications of Scale: Important to understand scaling issues, particularly when interpreting results from graphs and making judgments on statistical significance.
10.6 What’s Next?
Move from comparing two groups to more complex analysis such as ANOVA in the next chapter for scenarios with multiple subgroups, expanding the model utilized in this chapter.
10.7 Exercises
Conceptual Questions: Examine hypotheses on expenditures based on class status.
Statistical Application Problems: Identify educational attainment vs. COVID-19 cases using practical R commands and analyses to illustrate learning.