The shift among white voters with some college was less dramatic—from a 4-point Republican advantage in to a point advantage in —but still represents a large change in such a short time period. The second category is made up of individuals who have traditionally identified as or voted for Republicans but voted for Obama in While people have grown accustomed to thinking about those who voted for Obama in but not Clinton in as Democratic defectors, the reality is that some portion of these voters were really Republican defectors in and have now returned to their customary voting behavior.
Taken together, there is substantial reason to think that a good portion of these white non-college-educated voters were unlikely to vote Democratic in While additional resources aimed at reaching out might have resulted in smaller shifts, it seems unlikely that any discrete intervention in would have fully recreated the margins observed in Let us assume for a moment that the difficulties described above are real.
What if Democrats, even with a concerted effort, had only been able to split the difference between the and support rates of white non-college-educated voters? Under these conditions, Clinton still would have taken Michigan, Wisconsin, and Pennsylvania, and therefore won the election. The size of these wins is obviously smaller, but the narrowest is still a 1. Our simulation predicts that Clinton still would have carried these three Rust Belt states, with Pennsylvania now going Democratic by a very narrow 0.
On its own, Latino support returning to its levels would not have altered the outcome of the election or the outcome of any state. The simulation clearly has the biggest effect in Florida, but results in no Electoral College change. In many ways was about the U. No demographic exemplified that more than Latino voters. According to our analysis, the percent voting Democratic declined 3.
While no one has seriously argued that the election hinged on changes in Latino voting behavior, we found this simulation worthwhile to explore given the rising influence and unique electoral features of this group.
As expected, our results suggest that a return to levels of support would not have resulted in a win for Clinton. Geographically, Latinos can almost be considered an inverse image of white non-college-educated voters. While the latter had an outsized influence on the election because of their geographic distribution, the former punches below its weight. Latino voters tend to be concentrated in a relatively small number of counties in the country and those counties tend to be located in non-swing states.
The three states with the largest percentage of Latino voters—New Mexico, California, and Texas—were uncontested in and probably will be for at least several more presidential cycles. Of the next three—Arizona, Florida, and Nevada—only Florida is a true swing state, at least for now. That said, our simulation shows that even in this relatively high population state, Latino voters shifting back to their support levels would not have closed the gap for Clinton. It still would have missed the mark by about 5, votes.
State-level demographic changes were not pivotal in , but they did create conditions that were generally more favorable for Clinton. Absent any changes in the eligible voter population, several states that Trump won narrowly would have been much safer for him.
The simulation results in no Electoral College change. Demographics may not be destiny, but in the short term it is reasonable to quantify the effects of demographic changes on election outcomes. Our fourth simulation measures this very thing: What would the election look like if there had not been any demographic changes in the past four years?
We held turnout and support rates constant, but fed them into the demographics that were observed back in The effect is nearly universal—the demographic changes observed since have created an electoral landscape that is slightly more favorable to Democrats.
Had the population somehow remained unchanged during this time period, we expect Clinton would have won the national vote by 1. While these changes did not prove pivotal in any state, the estimated effect is still rather substantial. In our seven states of interest, this simulated population stability would have resulted in margins even more favorable for Trump.
In Michigan, Wisconson, and Pennsylvania Trump could have achieved margins that were 0. Although these states were not close, Ohio and Iowa would have seen a similarly sized 0. However, North Carolina and Florida, two states undergoing relatively rapid demographic shifts, were the most affected by changes to their eligible voter population.
Absent these changes, Trump would have expanded his win by 0. Broadening our horizons slightly, a number of states were significantly more Democratic in than they would have been given a stable population.
Georgia and Maryland were even more extreme, with margins around 1. In fact, only two places were Republican-advantaged as a result of demographic changes since Washington, D. Both have black populations that are shrinking as a share of eligible voters while less Democratic-leaning voters who are white and college-educated, Latino, and Asian or other race are growing. The demographic churn within these states creates a unique scenario: populations that are simultaneously becoming more racially diverse and less Democratic.
The findings from these data and simulations suggest that many of the existing intra-Democratic Party debates about the path forward have missed the mark. Rather than deciding whether to focus on 1 increasing turnout and mobilization of communities of color, a key component of the Democratic base, or 2 renewing efforts to persuade and win back some segment of white non-college-educated voters and to increase inroads among the white college-educated population, Democrats would clearly benefit from pursuing a political strategy capable of doing both.
If black turnout and support rates in had matched levels, Democrats would have held Florida, Michigan, Pennsylvania, and Wisconsin and flipped North Carolina, for a to Electoral College victory. So increasing engagement, mobilization, and representation of people of color must remain an important and sustained goal of Democrats. They cannot expect to win and expand their representation in other offices without the full engagement and participation of voters who are black, Latino, and Asian American or other race.
Given the fact that the white non-college-educated voting population is almost four times larger as a share of the electorate than is the black voting population, it is critical for Democrats to attract more support from the white non-college-educated voting bloc—even just reducing the deficit to something more manageable, as Obama did in and Likewise, the apparent shift to third-party voting and potential disengagement among younger voters must be considered carefully if Democrats are to make gains against Trump and Republicans in President Trump can conceivably reconstruct his primarily white coalition from with very few changes and still eke out a narrow Electoral College victory in But this assumes that Democrats do little to either increase the turnout of voters of color or to make inroads with disaffected white Trump voters, particularly Obama-Trump voters.
Alternatively, both Trump and Republicans could expand their electoral advantages among white voters by focusing and delivering on their economic promises on infrastructure, jobs, and wages and doing more to help people with health care. Given the trajectory of the current administration, this seems unlikely and could actually lead to a schism and third-party split among Republicans. Rob Griffin is the director of quantitative analysis at the Center for American Progress.
The authors would like to thank Lauren Vicary, Emily Haynes, Will Beaudouin, Steve Bonitatibus, and Chester Hawkins for their excellent editorial and graphic design work on this report. For this project we developed original turnout and support estimates by combining a multitude of publicly available data sources. We did this in order to deal with what we believe are systematic problems with some of the most widely available and widely cited pieces of data about elections.
One of the underappreciated problems in the world of election analysis is that some of the most reliable sources of data available on demographics, turnout, and support do not play very well together. For example, if we combine some of the best data we have on demographics with the best data we have on turnout, we find that they vary from the actual levels of turnout observed on Election Day. These estimates are fully integrated with one another and, when combined, recreate the election results observed in and Below is a more detailed description of how each was created.
We started off our process by collecting detailed demographic data at the county level from the U. The goal of this process was to produce reasonable estimates about the composition of eligible voters within a given county.
Specifically, we wanted to know how many eligible voters in each county fell into each our 32 demographic groups. For example, data on the race and age distribution as well as data on the age and education level distribution within a county are available separately. To overcome this problem we employed a two-stage estimation process. We then used iterative proportional fitting IPF to make these various pieces of data that are available line up with one another.
IPF is a form of adjustment that allowed us to make individual group counts—for example, the number of eligible voters in a county who are black, 18—29 years old, and have a college degree—line up with known population margins—for example, the number of eligible voters who are black and have a college degree, the number of eligible voters who are age 18—29 and have a college degree, and the number of eligible voters who are black and age 18— At this point in the process we had estimates on the eligible voter composition of each county, but there were several notable problems.
First, the use of the 5-year ACS was necessary in order to get estimates for every county in the United States, but it provides a somewhat blurry image of the year in question. Data from the 5-year ACS are an amalgamation of data from —, while data are from — In short, the ACS provides the necessary coverage but at the expense of giving us an accurate picture of the population as it existed in the year in question.
Second, the IPF process tends to spread certain characteristics—say, citizenship—somewhat indiscriminately across groups so long as the totals line up with other margins.
This is particularly problematic for something like education groups where—outside of the non-Hispanics white population—we see different rates of citizenship. Third, the IPF process inevitably generates estimates that are logically consistent within a county given the margins that have been provided, but does not collectively add up to the number of people one can expect to belong to a given group in a state.
To address all three problems we included an additional corrective step. Using the individual-level data from the and 1-year American Community Survey, we could accurately estimate the real state-level race, age, and education level composition of eligible voters. Logically, the numbers of eligible voters who fall into our 32 groups in the counties must add up to the number observed at the state level.
We once again employed IPF to make the frequencies in the counties collectively line up with the frequencies at the state level. These were used as our final estimates for eligible voter composition in each state. The process of creating county-level and turnout rates for each of our 32 demographic groups began by generating state-level estimates for these groups. Using data from the and November Supplement of the Current Population Survey, or CPS, we ran cross-nested multilevel models that estimated the turnout rate for each year, state, race, age, and education level group represented in the data.
We then fed those state-level turnout estimates into the eligible voter counts we generated in the previous step. This provided us with an initial estimate of how many people turned out to vote in a particular county in each year. At this point the difficulties we previously described became apparent—the estimated number of voters from a given county will inevitably deviate from the real number who voted. Once again, we employed IPF at the county level to force these counts to match up with one another, increasing or decreasing the turnout rates for our 32 groups until the two aggregate vote counts aligned.
Instead of treating the numbers as completely accurate, we view this process as something that helps us generate more precise state-level estimates. Namely, this process takes advantage of geographic segregation at the county level to selectively adjust turnout rates between demographic groups rather than applying a blanket correction at the state level.
Looking at Figure A1—which shows the share of eligible voters in each county who are white and do not have a college degree—we can see that there are some places where more than 80 percent of the population falls into that demographic category.
To the extent that our 32 demographic groups are non-randomly distributed across a state, this process will selectively push and pull their turnout rates. While the estimates within any given place may be off, we believe this discriminatory adjustment provides a better state-level picture. Combining our eligible voter estimates with our turnout rates, we could generate counts for the number of individuals in each county who voted and belonged to one of our 32 demographic groups.
The state-level compositions we reported throughout the paper are simply aggregations of these county counts. We feel that these estimates are superior to the ones typically reported from the November Supplement of the Current Population Survey for two reasons.
First, the multilevel modeling previously mentioned helps produce better estimates for small populations across the country. Second, when compared to the ACS the CPS would appear to systematically underrepresent the number of eligible voters in the population who are white and do not have a college degree.
As can be seen in Table A1—which compares the composition estimates from the CPS November Supplement and the 1-year ACS—white non-college-educated citizens age 18 or older are underrepresented by 1.
Given the superior sample size of the ACS, we believe it provides a more accurate picture of the eligible voter population, particularly at the state level. Assuming it is more accurate, post-stratifying our turnout estimates from the CPS onto the ACS eligible voter counts should provide a more accurate picture of the electorate. The process of creating county-level and Democratic and Republican support rates for each of our 32 demographic groups began by generating state-level estimates for these groups.
Using publicly available data from the American National Election Study and the Cooperative Congressional Election Study in and , as well as one of the post-election surveys from by Center for American Progress, we ran cross-nested multilevel models that estimate the turnout rate for each year, state, race, age, and education group represented in the data. We then fed those state-level support estimates into the voter counts we generated in the previous step.
This provided us with an initial estimate of how many people voted Democratic, Republican, and third party in a particular county in each year. Once again, the difficulties we described became apparent—the estimated number of Democratic, Republican, and third-party votes from a given county will inevitably deviate from the real election results. We employed IPF at the county level to force these counts to match up with one another, increasing or decreasing the support rates for our 32 groups until the aggregate vote counts aligned.
Instead of treating the numbers as completely accurate, we view this process as something that helps us generate more precise state-level estimates than previous methodologies. We see the strengths and weaknesses of this process in the same light as we previously described in the turnout explanation above.
Geographic segregation at the county level lets us selectively push and pull the support rates of our groups around rather than applying a blanket correction at a higher geographic level.
The estimates within any given place may be off, but we believe this discriminatory adjustment provides a better state-level picture. For this simulation we used the eligible voter composition within each county, but substituted the black turnout rates and party support rates with their counterparts. All other racial and educational groups were assigned their original turnout and support levels. Vote counts were then aggregated to the state level and reported. For this simulation we used the eligible voter composition within each county, but substituted the white, non-college-educated party support rates with their counterparts.
For this simulation we used the eligible voter composition within each county, but substituted the Latino party support rates with their counterparts. For this simulation we used the turnout rates and party support rates for every racial and educational group, but substituted the eligible voter composition in each county with its counterpart.
Ruy Teixeira , John Halpin. Ruy Teixeira. In this article. The product of this analysis is the following for each of those 32 groups: County-level estimates of eligible voter composition County-level turnout estimates County-level estimates of voter composition County-level party support estimates These estimates are fully integrated with one another and, when combined, recreate the elections results observed in and How much did differential turnout rates between white voters, including those who are college educated and those who are not college educated, and voters of color, including those who are black, Latino, and Asian American or other race affect the outcome of the election?
What exactly happened with the white vote, especially the white college-educated and white non-college-educated vote? How large is this latter group of voters compared to others?
Was there a big surge in support among white non-college-educated voters for Donald Trump, or not? How well did Hillary Clinton do with white college-educated voters compared to President Barack Obama? What exactly happened with the racial minority vote? Did Republicans do better or worse with black, Latino, and Asian American or other race voters? How did these turnout and support dynamics by group influence the outcomes in key Electoral College states such as Florida, Michigan, North Carolina, Ohio, Pennsylvania, and Wisconsin?
If black turnout and support rates in had been equal to black turnout and support rates in , what would the results have looked like in ? What about Latino margins? If white non-college-educated support for Democrats had been equal to white non-college-educated support for President Obama in , what would the results have looked like in ? The popularity of the Conservatives among BME voters fell by four points on the previous general election.
Turnout was at a year high, boosted by young people and BME voters. More than half of those aged turned out to vote, an increase of 16 percentage points on Turnout among BME voters also increased six points. Of those who went out to vote, but did not vote in the general election or the EU referendum, most voted for Labour. In a year when immigration played a central role in the presidential campaign, turnout among naturalized-citizen voters those who were immigrants born in another country who have naturalized to become U.
Overall, the voter turnout rate among foreign-born citizens trailed that of U. In , turnout among Asian naturalized citizens was Among Hispanics, naturalized-citizen turnout was Leading up to the election, the overall eligible voting population was the most racially and ethnically diverse ever. However, whites made up Meanwhile, blacks made up Hispanics have accounted for a growing share of the electorate for decades, and this trend continued in , when they made up 9. Asians made up 3. Millennials those ages 20 to 35 in had a Their turnout rate increased across racial and ethnic groups, with the exception of black Millennials, This increase in the Millennial voter turnout rate is not only because the generation has grown older older voters vote at higher rates than younger voters , but also due to a higher turnout rate among its youngest members: Generation X those ages 36 to 51 in turnout was By contrast, the voter turnout rate among older generations was flat.
Turnout for Baby Boomers those ages 52 to 70 was Note: Item No. The rate increased among white women, to
0コメント