Rhode Island Swings by Town & Correlation with Income

[Meatball Ron]

I’ve spent the last few hours playing around with the results of the 2012 and 2016 presidential election results in Rhode Island by town, using data from the state’s Board of Elections website. Hopefully someone will find my findings to be interesting or useful. Note that I use the word “town” to mean any official municipality, whether its actually a town or a city.

In 2012, Obama beat Romney in Rhode Island by a margin of 27.5 percentage points, but in 2016, Clinton beat Trump in the state by only 15.51 percentage points. This mean the state swung toward Trump by 11.99 percentage points, which is much higher than the national popular vote swing from 2012 to 2016 of only 1.8. This makes sense, as Rhode Island is considered to be a heavily elastic state.

However, at the town level, the swing was FAR from uniform— here’s where it gets interesting. At one extreme, you have Burriville, the northeastern most town in the state, which Obama won by 9 percentage points and Trump won by 22.2 percentage points, meaning a swing of 31.2 percentage points toward Trump. The results are similar in other western and central Providence County towns (the more urban, eastern parts went heavily for Hillary), as well as the western portion of Kent County (Coventry swung to Trump by 25.7 percentage points, West Warwick by 24.2 percentage points, and West Greenwich by 21.2 percentage points).

At the opposite extreme, wealthy Barrington swung toward Clinton by 16.1 percentage points, as she managed to win the state by 34.2 percentage points while Obama only won it by 18.1. Oddly enough, the town with the next largest swing toward Clinton was East Greenwich, a town that she won by 12 percentage points even though Obama lost it by 1.5 percentage points. I say “oddly” because this town borders the western Kent County towns I mentioned above that swung hard toward Trump.

Intuitively, it makes sense that wealthy, white, social moderates in towns like Barrington and East Greenwich preferred Romney to Trump and Clinton to Obama, while poorer white voters in the suburban parts of Providence County as well as the rural-ish towns of Washington County such as Richmond and Hopkinton heavily preferred Trump to Romney. These findings are not particularly novel, as they align with our understandings of today’s party coalitions as well as the national narrative about white working class voters having swung hard to Trump.

However, I wanted to statistically test this relationship between a Rhode Island town’s swing toward Trump and its median income to see how robust it truly was. Using a simple regression model, I found that, on average, as a town’s median income decreased by $1000, it swung 1.2 percentage points harder toward Trump. This relationship is undoubtedly statistically significant, with a p-value of 0.00002403.

One complicating factor, though, is race. The vast majority of Rhode Island towns are over 90% white, including all the towns I’ve mentioned so far. This is not the case, however, for the four major cities in Rhode Island: Providence (37.6% non-Hispanic white, swung 1.6 percentage points toward Trump), Central Falls (40% non-Hispanic white, swung 7.5 percentage points toward Trump), Pawtucket (59.5% non-Hispanic white, swung 13 percentage points toward Trump), and Woonsocket (71.3% non-Hispanic white, swung 24.1 percentage points toward Trump). I was worried that factoring in race might weaken this relationship, so I ran another regression where I controlled for race, and luckily this was not the case. In fact, it made the relationship stronger. With the variable of race (measured by percentage of a town’s residents who are white) factored in as a control, I found that, on average, as a town’s median income decreased by $1000, it swung 2.08 percentage points toward Trump. This relationship was also statistically significant with a p-value of practically zero. It makes sense that the relationship was even stronger when I controlled for race, because as you’ll see on the graph below, these cities with a large percentage of non-white residents were the outliers in this correlation, in that they have relatively low median incomes but did not swing as hard toward Trump as other areas of the state (with the exception of Woonsocket).



This graph should be pretty self-explanatory, but it shows the correlation that I've described. Each point on the scatterplot represents a RI town; the size of the dot represents its percentage of non-white residents and the color of a dot represents how it voted in 2012 and 2016 (see the legend on the side). Note the three major outliers in the center left -- the large and diverse cities of Providence, Pawtucket, and Central Falls. These cities have very low median incomes but did not shift as heavily toward Trump, largely due to their high percentage of non-white residents. Other than that, the relationship largely holds. Note the single red dot (Romney/Clinton), East Greenwich, which is both the wealthiest town in the state and the town that swung hardest toward Clinton. It's very much an outlier within Kent County, the only county that went for Trump overall, whose other towns can be found in purple and green at the opposite end of the graph.

Hope this was interesting to people! If you'd like the raw data, feel free to send me a message.

[Meatball Ron]

A couple updates:

I standardized the data, so that all election data comes from the Board of Elections website and all race/income data comes from the Census Bureau. Previously, I was using a mix of the BoE site, the Census, CNN and NBC's election results, and Wikipedia, which I realized would lead to some inconsistencies.

This made the relationships slightly less substantively significant. In the first model (no control variables), the new coefficient is 0.77, meaning that on average, as a town's per capita income decreases by $1,000 its swing toward Trump increases by 0.77 percentage points. In the second, more accurately representative model (control variable for race), the new coefficient is 1.46.

I also updated the scatterplot to reflect these data changes, and perhaps most excitingly, I added the names of individual towns to their corresponding points on the plot.

[Meatball Ron]

Hey RecoveringDem,

I just wanted to jump in and say this is some pretty awesome work, and love where you are going with this project!

Haven't had a chance to really read through and totally digest, since it's both a workweek for me as well as state where despite having family members out there in RI, haven't really looked at too much politically over the years.

Anyways--- just wanted to say I'll need to take more of a look on my upcoming four day weekend, since, there is a ton of data packed in there, and helps show where the '12 > '16 swings happened in RI using a graphical format, as well as potentially indicating direct correlations that might be portable to other places in New England.

Thinking it would also be interesting to incorporate an Ancestry element chart, mainly because all of the discussion regarding Ethnic-European populations in RI (Italian-Americans, Irish-Americans, Portuguese, and even/especially (?) French/ French Canadian).

Elsewhere I did a few posts regarding the "Poorest Voters" in the "Poorest Counties" by State, and it seems that in New England, there might well be a strong connection between those that identify as French/ French-Canadian in terms of Ancestry, and both Poverty and dramatic swings towards Trump between '12 and ;16.

https://uselectionatlas.org/FORUM/index.php?topic=267947.msg5741987#msg5741987

What happened in Woonsocket, RI for example that is 41% French/ French Canadian and is essentially a manufacturing town that went 66% Obama in '12 on only 50% Clinton in '16?

Also, one of the things that I had issues with dealing with RI, is that it is way easy to get "City" / "Township" data, but trying to obtain data drilled down to individual precincts is extremely difficult.... It appears to be a major issue in particular in New England, but obviously if we can't drill down into extensive detail in a place like Providence, for example, it's going to be much harder to analyze RI, other than going through a "City based format", which inherently loses granularity, especially when trying to examine voting swings among various Social-Demographic populations.

I would also imagine that  '04>'08>'12>'16 graphs might be interesting to look at as well to assess the various shifts in support, since it doesn't appear that all of these Trump '16 voters hadn't voted Democratic previously (Or what does the data say???).

Great work--- effort posts/ thread.  

Thanks for the kind words! I joined this forum hoping there'd be lots of statistical analysis of election data. But it seems people are more interested in brainstorming bizarre pairings of obscure potential 2020 candidates. It's all good though

[Meatball Ron]

So---- on the subject of RI--- you didn't answer the question in my response regarding Woonsocket and the French/ French Canadian vote in RI....   

Yeah, sorry about that! I honestly just don't know too much about the different white ethnic groups and their voting patterns. Woonsocket's swing sort of makes sense to me-- it swung harder than the other major cities (Providence, Central Falls, Pawtucket) because it has less minority voters than them, but not as hard as the other low-income white towns because it has more minority voters than them. I bet there's more to the story when you look at the individual ethnic groups, but I'm not sure if the data is out there. I'll look into it!