The Most Interesting Survey in the World: the General Social Survey (1972 – 2012) | Part 1

The General Social Survey 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.

Until now, exploring this dataset required painstaking downloads, codebooks, and installed software. Now anyone can play around with it online in Statwing.

We took a subset of the American GSS and created a quiz based entirely off of novel findings from our analysis of that data., Enjoy!

Americans were asked whether the government should spend more, the same, or less on a variety of government programs.
About which area do men and women disagree most strongly?

Wrong

Wrong

Correct. Space!

Wrong

It’s official: men love space! Across the whole dataset (1972-2012), 20% of men thought we should spend more on space, versus less than 10% of women.

Men love space

In this crosstab, a screenshot from Statwing, every column sums to 100%. So, 35% of men think we spend too much on space, versus about half of women.
Those arrows indicate that a cell is statistically significantly high or low.

We didn’t look into how that level of disagreement has changed over time. If you’re curious, you could see the above analysis in Statwing, then filter the data for various survey years.

You’ve gotten out of answers correct so far.

In the 1970s, 57% of strong Democrats thought we should spend more on the environment, versus 47% of strong Republicans, a gap of 10%.
How large was the gap in the last few years?

Wrong

Wrong

Correct

Across the last two surveys, 73% of strong Democrats thought we should spend more on the environment, vs. 32% of strong Republicans.

Polarization

Note that in this case, each row sums to 100%.
The data does back up the perception that American politics have become more polarized. One sees similarly increased gaps on quite a few issues.

One thing we didn’t explore: what was the trajectory of this increased polarization? Did it shoot up suddenly in the last decade or two, or has there been a smooth upward slope? You can see the above table in Statwing and filter for different years to investigate further.

You’ve gotten out of questions correct so far.

In 1972, 75% of the random sample were married. What percent of a new random sample in 2012 were married?

Wrong. It’s actually 45%. Pretty crazy, eh?

Wrong. It’s actually 45%. Pretty crazy, eh?

Correct

Wrong. It’s actually 45%. But you’re not far off.

Marital status

In this chart, every row sums to 100%. So, for example, in 1972 75% of respondents were married.
We all know that marriage has declined in the U.S., but the numbers are pretty stark.

The analytically inclined might wonder how much of the increase in “Never Married” is driven by people marrying less versus marrying later. To find out, go to the above analysis in Statwing then filter for different age ranges.

You’ve gotten out of answers correct so far.

Which group is more likely to live in the same city that they lived in when they were 16?

Wrong

Correct

Wrong

A bit more than half of black folks live in the same city they grew up in, as opposed to less than 40% of white folks. Not a huge difference, but interesting. The effect remains if you control for age and only look at recent years, too.

Race and mobility

col100. More than half of black respondents reported living in the same city they lived in when they were 16.

We didn’t look into the reasons behind this. Different races do come from differently sized locations, and have different average incomes, and there’s a few other reasonable hypotheses that could explain this. If you go to the above analysis in Statwing, you’ll probably want to explore with variables like “Age”, “Race”, “Population Density when 16″, “Income when 16″, and “Year of Survey”.

You’ve gotten out of answers correct so far.

Which group is most disliked in the U.S.?

Wrong

Wrong

Wrong

Correct

Wrong

The GSS asks three questions about these groups: should they be allowed to give a speech in town, teach at a university, or have a book they authored in the local public library.

Unpopular Groups

Click through the image to explore in Statwing.

Probably the most interesting thing about this analysis was the extent to which stated attitudes towards different groups have changed over time.
Homosexual Speech

Stated attitudes towards homosexual folks changed pretty consistently in the last 30 years.
Some of the other groups have seen similar changes, others have not.

You can explore in Statwing how opinions towards the other groups have changed.

You’ve gotten out of questions correct so far.

In 1978, about 50% of surveyed American men had been on active military duty at some point in their lives. In 1988 about 40% had been on duty.
What was that number in 2012?

Wrong

Wrong

Wrong

Correct

This question is tricky because you’d have to account for survey respondents having lived through multiple wars—in the 1970s many respondents would have lived through WWII and the Korean War, for example.

Active Military Duty

This data is filtered for men only. It’d also be interesting to filter for, say, men aged 18-22, or 50-60.

It’d likely be interesting to analyze that same data while filtering for respondents of various ages (e.g., proportion of 20-25 year olds who had served). Here’s the above analysis in Statwing to play around with.

Thanks for playing

You got out of answers correct. If you’d guessed randomly, you’d have had a 85% chance of getting one right. So. Yeah. You have a keen eye for exactly what isn’t happening in America. Not ideal. But, hey, such is life. That’ll do just fine. Impressive. Ummmm. Wow. You should probably have a newspaper column somewhere about American culture.

Tweet your quiz results

Discussion on Hacker News

We’ll be writing a lot more posts based off of this dataset. It’s absolutely bountiful. Subscribe via the little box to the left if you want more like this. Or play around with the data yourself–here’s the list of variables available in the dataset.

Plug: Statwing makes it easy for anyone to analyze data like a data scientist. If you have an app that produces or collects data for your users, you can use our API to build an Export to Statwing link so your users can get more value out of their data.

Notes

Learn more here.

Findings are “novel” in the sense that we didn’t go look up other folks’ papers, these quiz results are based off of our own analyses.

Methodological notes:

  • Of the many modules, we grabbed “Case Identification and Year”, “Respondent Background Variables”, “Personal and Family Information”, and “Attitudinal Measures – National Problems” bulk download part of Berkeley’s GSS site (that’s roughly 10% of the core dataset).
  • We selected almost all the variables that (1) were asked across many different years, typically including 2012, (2) seemed at least somewhat interesting, and (3) were readily interpretable. For example, respondents were asked for a few pieces of information about every other member of the household; we left those out. While that could yield some interesting analyses, it’d take some really specific data-reshaping and analyses to get value out of it, relative to everything else. We ended up with the vast majority of the data.
  • We only used every other year of the data.
  • We excluded “Don’t know” answers from the dataset. Sometimes those can be interesting, but often they’re just annoying.
  • 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 randomly filtered out some cases; the 1982 survey oversampled black respondents, so we also randomly eliminated some of those cases.
  • 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 still very valid even if the absolute numbers are slightly different from a perfect poll (not that such a thing exists).

In 1972 the GSS asked only if respondents were white, black, or other. In more recent times they adopted the same line of questioning as the modern census, which allows for more fine-grained analysis, but they continue to back out the 1972 result. We use the 1972 classification-type here to maintain the time series (though it doesn’t matter for this particular question). We understand that this is suboptimal, and apologize if this comes off as a bit insensitive. More details.

“Atheist” is perhaps a bit unfair; technically the question asked about “somebody who is against churches and religion”.

Sorry, comments are closed for this post.