The Most Vulnerable Georgia Districts — Part Two
If you haven’t read the earlier parts of this series, then please find them here:
We have our targets from the HVI analysis, but is there anything that analysis misses?
The biggest criticism I have of the HVI is that it assumes everything will be the same as it was in the previous elections: demographics, turnout, registration, etc. I think it is a good starting point, but if anything has changed, then we need to account for that when targeting districts. So let’s take a look.
What can a demographics model tell us?
Previous results are a place to start, but what about predicting future results? Are there districts that could switch parties if a significant group moves in our out of the district? What if a certain group turns out to vote in higher percentages? What if a certain group starts voting for a different party?
The answers to these questions can be predicted based on public information. We can highlight demographic changes and deficiencies in voter turnout or registration rates. We can predict how each district would vote based on different levels of turnout or changes in the partisan tilt for each demographic group. If we can predict the district vote based on different benchmarks, then we know where to target and what to expect on election day.
Building A Simple Demographics Model
The idea behind the model itself is simple:
Democratic votes = Registered voters * % Turnout * % Voting for Democrats
Republican votes = Registered voters * % Turnout * % Voting for Republicans
We just need to gather the data and determine how to drill down into that data.
Registration and Turnout Data
The Georgia Secretary of State’s office posts registration and turnout data split by State Senate and House district.
The registration information is current as of September 1, 2016. Historical registration information is included with the turnout data. Unfortunately, the turnout data for 2014 is only split by county. The turnout will lean towards the turnout in 2012 and 2008 rather than a mid-term election, so the lack of 2014 data is only a minimal loss.
The 2008 turnout information is broken down by district, but of course, the 2008 districts are not the same as districts today. To account for this, I took the average change from 2012 to 2008 and applied it district-by-district to the 2012 numbers.
Demographic Splits
While there a few options, I’ve decided to break down the population and registration data into the following groups:
- College-educated White
- Non-college-educated White
- Black
- Hispanic/Latino
- Asian (Includes Native Hawaiian and Pacific Islander)
- Native Americans
- Other
I’ve estimated the number of college-educated whites based on the population numbers from the American Community Survey’s 2014 numbers. It’s not perfect, but the partisan lean of college-educated whites and non-college-educated whites is significant enough to be included even if by an estimation.
Ideally we could split each of these groups even further by age and sex, but unfortunately, there would be a significant amount of error in those estimations. Registration and turnout information is split by sex but not by age. It would be easy enough to split the data by sex, but the issue comes with estimating the vote share down to that level. Current polling data doesn’t break down to that level of detail, so everything would be extrapolated from the totals.
It would be an even more daunting task to use the age data. The ACS’s voting age population numbers are not broken down by both race and age, so we would have to once again extrapolate from the estimated percentages of each age group within each district. I’ll be working on trying to include this data in further models, but for now, the current groups will have to do.
Partisan Voting Share and a Baseline
This is where things get interesting. We need to estimate what percentage of each group will vote for each party. To predict future results, I needed a baseline to work from. The first step here was to estimate the voting share in 2012 based on registration and turnout in each district. The presidential voting result within each district is a fair estimate of the partisan voting share, so if I could match the model’s voting share to the presidential results, then I knew the model should fairly predict the results when using alternative voting share and turnout rates. I started with estimates from FiveThirtyEight.
After making some adjustments, I found a voting share that accounted for the total voting share across the state as well as in most districts. Most districts were estimated within a few percentage points of the actual results, but a few districts didn’t even come close. The districts with the greatest error were districts that had a heavy skew towards a certain age group or sex. For example, females and young voters are skewing towards the Democrats, so districts that have a greater than average percentage of those groups are underestimating the Democratic share of the vote.
As I mentioned, I will be working on the model between now and the election to help account for those and create a better estimate as time goes on. However, to work with the model now, I adjusted each district’s estimate to match the presidential total from 2012. For example, if an estimate for a district was 5% above the actual party share of the vote, then I reduced the estimate for 5%. These adjustments are applied regardless of whether I’m using the 2012 estimate or any other estimate of the voting share. The errors and adjustments are in the table below.
Vulnerability 2.0
Starting with the 2012 baseline, there are a variety of conceivable scenarios:
- Use 2012 turnout/voting share to highlight only demographic changes
- Use average turnout and an average of the polling data to predict outcomes based on the current political climate
- Estimate the outcome if every person of the voting age population voted to highlight registration deficiencies
- Estimate the outcome if every registered person voted to highlight turnout deficiencies
- Use only presidential polling data and 2008 turnout to estimate the best case scenario
- Use only senate polling data and 2012 turnout to estimate the worst case scenario
Each of these scenarios will provide an interesting look at the state of the race, but let’s start with the first option to provide the most basic changes in the demographics. This will mimic the 2012 conditions (same turnout and same estimated partisan vote share) using the registration data as of September 1. I’ll start by categorizing the districts as we did in Part One, and then we can take a closer look at the borderline Category 1/2 districts.
We will use the same criteria for the three targeting categories based on presidential margin:
- Category 1 = >-2.5% margin
- Category 1 or 2 = >-7.5% margin
- Category 2 = >-15% margin
- Category 3 = Category 1 for the Republicans
I will also list the HVI Category for each district in parentheses.
Senate Targets
Category 1 —
>-2.5% margin:
- SD-43 (1)
No district meets the alternative criteria for Category 1
Category 2 —
>-15% margin:
- SD-6 (1)
- SD-40 (1)
- SD-48 (None)
- SD-23 (1)
Category 3 —
No districts meet the criteria for Category 3
Changes from HVI —
The only addition from the HVI analysis is SD-48 into Category 2. The estimate here is a 12.42% deficit for the Democrats, but it is a district to keep in mind as we look to 2018 and beyond.
SD-8 was #4 in the HVI analysis but is only #7 based on the demographics here. This is just another example of the uncompetitive nature of these senate districts. In later posts, I will analyze these districts individually to show where there are advantages for the Democrats. But the House still presents a better opportunity for gains.
House Targets
Category 1 —
>-2.5% margin:
- HD-151 (1)
- HD-111 (1)
- HD-105 (1)
- HD-101 (1 or 2)
- HD-145 (1)
- HD-138 (1)
- HD-107 (2)
>-7.5% margin:
- HD-106 (2)
- HD-40 (2)
Category 2 —
- HD-147 (2)
- HD-95 (1)
- HD-37 (2)
- HD-117 (2)
- HD-164 (None)
- HD-173 (2)
- HD-140 (2)
- HD-109 (None)
Category 3 —
>-2.5% margin:
- HD-80 (3)
>-7.5% margin:
- HD-81 (3)
- HD-132 (3)
Changes from HVI:
The changes in the demographics are easily seen in the House estimates. Everything shifts towards the Democrats. We add two new districts to Category 2, two from Category 2 to a potential Category 1, and two others from Category move entirely into Category 1.
Two districts (HD-51 and HD-79) aren’t included here, yet the HVI analysis had them as Category 2 districts. These were the final two included just above a -15% margin, but the current demographics have shifted that margin to just under -15%. They’re on the border and I’ll include them in the list of districts to examine further.
Do the current polls change the analysis?
The purpose of undertaking this analysis was to act as if I was the Georgia Democratic Party finding potential targets for the election. If however, we move beyond the initial query of the analysis and examine the current state of the race, then the story becomes a little more interesting.
Mimicking the 2012 voting share is a nice to way highlight the demographic changes, but the 2016 race is quite different than the 2012 race. Can we use more current data to show what opportunities are available to the Georgia Democrats this year? While not perfect, polling data gives us the best indication of the state of the race in Georgia.
To that end, I’ve made voting shares using both the presidential polling as well as the senate polling. Each of these races comes with its own baggage. The presidential numbers are skewed by Donald Trump. The Democratic share of the vote based on the presidential polls is much larger than, for example, the senate polls.
The polls for the Isakson-Barksdale race still favor Isakson for a variety of reasons. First, Isakson is a well-liked, incumbent Republican in a state that has no statewide elected Democrats. Second, Barksdale is still an unknown candidate. Isakson probably out-polls any Democrat right now, but by a lesser margin given a different candidate.
We can view these as the best case and worst case scenarios for the Democrats. Best case = down-ballot candidates keep pace with the Clinton/Trump race. Worst case = down-ballot candidates buck the Clinton/Trump trend and are still dominated by the Republican candidates. Each scenario tells us something about the state of the race and the Party, so it’s only fair to include both. My gut tells me that the likely outcome for the General Assembly races will fall between the drag of Trump’s alt-right shadow campaign and the drag of Jim Barksdale’s might as well run a guy off the street campaign.
Instead of pushing towards the best case or worst case, I took the average of the two to balance out the Trump factor with the Isakson incumbency factor. For turnout, I also used an average of the 2008 and 2012 elections. With these averages, we can examine what I think is the most likely scenario for the outcomes in November.
Again, without taking a look at the individual candidates (that will come later), let’s look at how the categories shift with the current polling.
Senate Targets
Right off the bat, nothing will change for Category 3. SD-33 is now the closest race for a potential Republican flip, but it has an estimated margin of 21.93% in favor of the Democrats. But what about the Republican-held seats?
SD-6 and SD-40 both move into a positive margin for the Democrats right around 4%. Those along with SD-48 move solidly into Category 1.
Seven districts (56, 8, 17, 9, 1, 46, and 29) move inside of the 15% margin for Category 2. SD-56 changes drastically with about a 14 point increase from a -24.05% margin to -9.87%.
Using the polls, the Senate starts to look a little more competitive. There are now three districts in a definite toss-up category, with a few more in striking distance for a surprise victory in 2016 or a competitive election in 2018.
House Targets
Yet again, the Republicans gain no advantage from the current polling. In fact, HD-80, which held about a 13% margin for Romney, is now an estimated 1.66% margin in favor of the Democrats. Nothing else is close enough based on our Category 3 criteria. Now on to the Democrats.
With these numbers, it looks like the Democrats have a significant opportunity to shake things up.
Districts 40, 117, 95, 37, 79, 54, and 51 are now in Category 1 each with an increase of over 10%. HD-106 also moves into Category 1 albeit with a smaller increase.
Districts 147, 119, and 164 are now toss-ups within a 5% margin while districts 108, 102, 43, and 173 are now above the -7.5% cutoff.
Districts 48, 179, 35, 52, 50, 171, 110, 44, 130, 49, 158, 73 all move above the -15% mark for inclusion in Category 2.
This is a significant shift for the 2016 election with opportunities for future gains in 2018 and 2020. Using these averages there are now 36 districts combined in Categories 1 and 2. If by the end of 2020, all 36 districts flipped, then the Democrats would hold the house 97/83. Of course, this assumes similar results in 2018/2020 as in this election, but this is a start.
Final Target List
Compiling each of the methods so far, here is my final target list for the Georgia Senate and Georgia House broken down by category:
I included any district that made an appearance in any of the methods. I used the highest category between the methods. For example, I included the top five based on HVI as Category 1 districts. SD-8 was #4 on the original HVI list, but using the updated demographics data didn’t appear in either category. In my opinion, we need to target at least the top five most vulnerable districts, so I still included SD-8 as a Category 1 target.
I will take each of these districts and examine them using alternative scenarios (senate polling, 2012 numbers, increased/depressed turnout, increased registration, presidential polling, etc.). During this analysis, I will also profile the candidates and the individual race itself. By the end, we should have a good idea on the outcome of each targeted race. I’ll also attempt to update to give a more accurate estimation based on the most current polling data.
Final Thoughts on Vulnerability
When starting this examination of competitiveness and vulnerability, I wanted to take an expansive view of potential targets. The Democrats have significant ground to make up by 2020. If we are to take back at least one chamber in the General Assembly by the time redistricting occurs, then we will need to expand the map and win districts we haven’t imagined in a while. A narrow view will not do this. We need to cast a wide net and see what we can catch.
You can see on the target list above that I’ve included a checkbox for whether a Democratic candidate appears on the ballot. I’ll get into this further as I analyze each district, but you can see a significant issue here. 12 out of 24 Category 1 and 16 out of 24 Category 2 districts have no Democratic candidate on the ballot. Regardless of demographics, turnout, or registration numbers, the Republicans will hold those seats.
A rudimentary model can give us some hope for the future by exposing new vulnerable districts. The Trump drag is real, and, combined with the shifting demographics, more districts are now within striking distance.
If only there was a Democrat on the ballot.
— What other factors would you like to see accounted for in the model?
— What polling numbers do you think the General Assembly elections will skew towards?
— Following the POLITICS of today, for the generations of TOMORROW —
Politics for Tomorrow is a publication focused on progressive politics both nationally and in the State of Georgia.
Follow Politics for Tomorrow on Twitter here.
Follow Austin Wagner on Twitter here.
