Yesterday Gallup published poll results showing an all-time low of Americans Identifying as Republicans. We statistically analyzed data from the General Social Survey to identify which groups were leaving the Republican party. The answer? Moderates, liberals, the wealthy, the young, and the well-educated, but not women.,
Republicans have been defecting to “Independent” for 20 years
For reference, here’s Gallup’s chart:
The chart derived from GSS data is similar:
This trend is driven by moderates and liberals leaving the party
Between 1990 and 2012, there’s no statistically significant trend of conservatives moving away from the Republican party. But during that same period, every year on average 0.40% of moderates and 0.49% of liberals (0.44% of the combined group) moved away from the Republican party.
Of moderates and liberals, those leaving the Republican party tend to be young, well-educated, and/or well-off
Women are not leaving the party any faster than men
There’s no indication, though, that women are leaving the party any faster than men.
The Global Social Survey (GSS) is the most frequently cited social science dataset (except for the U.S. Census). More than 14,000 academic papers, books, and dissertations are based on this biennial, 90-minute survey, conducted since 1972.
- Short version: the GSS is very sound, there’s no reason to be concerned about the robustness or validity of its results. Long version:
- Of the many modules, we grabbed “Case Identification and Year”, “Respondent Background Variables”, and “Personal and Family Information” from the bulk download part of Berkeley’s GSS site.
- We only used every other year of the data; the only exception is that there was no GSS in 1992, so we also included 1993’s data. This is mostly a function of how we like to present data in Statwing. This probably sacrificed a very small amount of statistical power, but all findings would be the same, since the GSS became biennial after 1994 anyway.
- About 10% of respondents were not asked or did not respond regarding their place on the Liberal/Conservative spectrum. They were roughly evenly distributed across the political parties, so it’s very unlikely that they would affect the analysis significantly. Only 0.6% of respondents were not asked or did not answer the question regarding their political party.
- The survey randomly selects households, not individuals. So individuals who live in households with many people are slightly unrepresented.
- The dataset slightly overrepresented women, so we filtered out a randomly selected set of cases so that every year’s Male:Female ratio was near .97:1, as per the American Community Survey; the 1982 survey oversampled black respondents, so we also randomly eliminated some of those cases.
- The GSS response options for the liberal to conservative spectrum were “Extremely liberal”, “Liberal”, “Slightly liberal”, “Moderate”, “Slightly conservative”, and so on. We condensed them because there weren’t important differences between degrees of liberalness or conservatism and because it made the chart cleaner.
- The GSS response options for political party were “Strong Democrat”, “Democrat, not strong”, “Democrat”, “Independent, lean Democrat”, “Independent”, “Independent, lean Republican”, and so on. We condensed them in the same fashion as Gallup.
- The resulting dataset can be downloaded here.
- In sum, you can’t say with 100% confidence “X% of Americans believe Y”, but you can get very close, and the trends and relationships are valid even if the absolute numbers are slightly different from a perfect poll. Of course, no poll is 100% perfect, but Gallup’s sampled 18,000 individuals and designed weights expressly for this purpose, so their absolute numbers are more likely to be correct than those of the Global Social Survey.
Insomuch as they disagree on absolute numbers, Gallup’s are the authoritative ones, since their sample size is much larger and their sample weighting is more finely tuned to exactly representing America’s absolute feelings—the GSS is authoritative for trends and relationships, like those explored here, not for absolute representativeness.
We fit a linear regression line, which is more robust than just looking at the first number and then the last and dividing by the number of years. The 0.31% number is plus or minus 0.04%. Every trendline mentioned in this piece is statistically significant with a p-value below 0.00001.
It’s possible that “new” 18 to 30-year old liberals and moderates are less likely to be Republicans than those graduating out of this range, so individuals aren’t changing their ID as much as new cohorts have different party identifications. This applies to all the other groups covered in this piece.
Note that while higher incomes are associated with greater education, both of these effects existed separately from one another. That is, given a Bachelor’s degree, people with a higher incomes were still more likely to leave the party, and vice versa.