The Cold Civil War Survey: Results

This post refers to a survey you can find here if you haven't already seen it. It received fewer responses than I had hoped, 149 before it scrolled off the front page, but the results were consistent enough across that sample to provide some insights. The shortage of numbers does pose some limits in determining statistical significance, as we'll see.

Network Breaking Behaviors

The hypothesis of this survey was that a sign of a Cold Civil War would be deliberate breaking of social and other networks along political alignments, reducing the strength of civil society and segregating what might become parties to a conflict. I have resorted the questions in decreasing order of occurrence of the behavior.

One caution on interpretation: These questions attempt to measure change in social networks for the self-selected respondents. They do not necessarily reflect starting or ending state of those networks in terms of political homogeneity. Put another way: Saying 'no' to a question says the respondent made no changes in that regard in the period from 2000-8. However, it does not distinguish whether that's because the individual didn't care about political issues that strongly, or whether the individual had already segregated his or her personal network by political beliefs before 2000. And so to the results:

Since November 2000 have you, due primarily to political differences:

Broken up with a significant other? 3%

There's an obvious trend from expendable commercial or virtual relationships downward to closer and more face-to-face connections. The one outlier to that pattern is moving away from political dissonance, with over twice as many respondents reporting this change as giving a neighbor the cold shoulder. Apparently the physical network is easier to break by wholesale.

The last four questions are dropped in further analysis, since they had so few positive responses that it wasn't possible to obtain statistically meaningful results. These behaviors are dramatic enough to make stories that may travel as gossip. Given that positive responses from the audience of this largely political blog are likely to exceed the general population, they seem to be fairly rare.

Does Party Matter?

Respondents reported party affiliation of 41% Republican, 20% Democrat, and 39% Neither. I included two issues questions as controls on the affiliation report, and to see if they would offer insight on the surveyed behaviors. The following distributions contain no surprises:

In the long run, 9/11 was:

Democrat: 53% "A declaration of war" 28% "A distraction" 3% "An inside job" 12% No response
Neither: 81% "War" 16% "Distraction" 2% "Inside" 2% No response
Republican: 97% "War" 3% "Distraction"

Dropping the troofer response for inadequate sample size, the correlation between party affiliation and response to 9/11 is statistically significant with P < .001.

The election of 2000 was:

Democrat: 47% "Stolen" 3% "Decided correctly" 50% "Just a mess"
Neither: 3% "Stolen" 52% "Decided correctly" 45% "Just a mess"
Republican: 2% "Stolen" 64% "Decided correctly" 34% "Just a mess"

The relationship between party affiliation and attitudes towards the 2000 election is significant, with P < .0001. Interestingly, on this issue the Neithers differ significantly from the Democrats (P < .0001), but not from the Republicans. The independents appear to have 'moved on' with respect to this issue. However, looking at the responses to the 9/11 question, the Neithers in that case do occupy a mid-ground between the Ds and Rs.

So, Does Party Really Matter?

Sometimes yes, sometimes no. Here are percentages by issue in the order given above, with respondents segregated by party:

Cancelled a subscription or stopped reading an author?
Democrat: 38%
Neither: 60%
Republican: 64%

Moved?
Democrat: 9%
Neither: 7%
Republican: 12%

Boycotting a company for political reasons is apparently itself a non-partisan choice. About half the respondents have done so, and affiliation makes no difference. Most of the remaining behaviors appear to show some variation by party, but only some of them are statistically significant.

Party affiliation matters in regards of dropping out of virtual communities (P = .0059) and dropping friendships (P = .028). There appear to be differences in other reported behaviors, but they fall below statistical significance due to sample size. Note again the caution with respect to interpretation: Democrats in this sample dropped virtual and face to face relationship significantly less frequently, but that doesn't necessarily mean their current networks are more or less homogeneous than Republicans or independents: We don't know the starting state of any of their networks.

The relationship of party affiliation to cancellation of subscriptions is marginally significant, but it turns out that more of the variation can be explained by sorting in a different fashion:

Cancelled a subscription or stopped reading an author?
2000 decided correctly: 68%
2000 just a mess: 52%
2000 stolen: 33%

The relationship between attitude toward the 2000 election and cancellation is significant with P = .0155. Looking at another behavior in the light of this issue also turns out to have some explanatory power:

Moved?
2000 decided correctly: 4%
2000 just a mess: 10%
2000 stolen: 28%

Significant with P = .01. Incorporating the partisan data from above, Democrats who believe the 2000 election was a fraud are significantly more likely to have chosen a place to live due to political issues, but less likely to have dropped subscriptions. (This slice of the data does not give further insight on the other behaviors. Slicing by attitude towards 9/11 didn't help either, due to the overwhelming fraction who called it 'war'.)

What's It Mean?

Based on this limited, self-selected sample it appears that there is self-segregation occurring at the margins. By that I mean that when the stakes are low, partisans will drop out of networks, ignore people with opposing views, and remove their monetary support from antithetical institutions. This drops off very quickly when the stakes are raised to close family or physical relationships. Cause for concern, but ignoring alone does not make a civil war, cold or hot. It's also likely no big surprise to those who have paid attention to earlier studies such as this network analysis of book purchasing patterns.

Getting further into speculation, there's some suggestion in the results that the most partisan of Democrats may have already segregated their personal networks before the relevant period, and some fraction of them then took the next step into physical separation after 2000. At the same time, a substantial fraction of Republicans and some independents followed them into sorting out the margins of their networks by political affinities. I want to be careful to label this as a hunch since the data don't directly tell homogeneity of networks.

Another open question is ongoing behavior. Is the ideological sorting suggested in some of these results a one time thing? Have the most politically sensitive moved themselves to their comfort zones, and that's the end? Or is this a feedback loop, with the partially segregated virtual and physical abodes becoming more philosophically uniform and strident, driving the less sensitive to also rearrange their circumstances? As Fazal Majid pointed out in comments to the original survey, Americans move on average once every seven years anyway. Even if political issues don't rank as the top concern for those moving, having it shifted up the prior list for relocators could have a substantial sorting effect over time.

Perhaps someone more ingenious than myself at surveys and network studies can figure out how to get at these questions. (Anyone out there is welcome to reuse or modify the current survey to add to these results.)

What do you think?

(Update: Statistical tests were performed with the chi-squared routines available on the Vassar College statistics website. Thanks! I've also imported the survey results into R for another look. That's mostly an exercise in trying a more powerful tool for possible follow-on surveys, but if anything new results I will amend this post.)