Midterm Election Poll: Illinois’s 6th District, Roskam vs. Casten (original) (raw)
Our poll shows a close race.
Our poll result is about what was expected. But remember: It’s just one poll, and we talked to only 497 people. Each candidate’s total could easily be five points different if we polled everyone in the district. And having a small sample is only one possible source of error.
This survey was conducted by The New York Times Upshot and Siena College.
Where we called:
Each dot shows one of the 53246 calls we made.
Vote choice: Dem. Rep. Don’t know Didn’t answer
To preserve privacy, exact addresses have been concealed. The locations shown here are approximate.
Explore the 2016 election in detail with this interactive map.
About the race
- Sean Casten is a scientist and a clean-energy businessman. 45% favorable rating; 34% unfavorable; 22% don’t know
Based on 497 interviews - Peter Roskam is the current representative and a former lawyer. He voted for the tax reform bill and to repeal and replace the Affordable Care Act. 47% favorable rating; 40% unfavorable; 13% don’t know
Based on 497 interviews - Illinois’s Sixth is emblematic of a type of district that could decide control of the House: a well-educated suburb that voted for Hillary Clinton but that usually votes Republican and has a strong incumbent against a political newcomer.
- (We polled this district from Sept. 4-6 and found a close race.)
- This district, in the western suburbs of Chicago, has a median income of nearly $100,000.
- Mr. Roskam won by 18 points in 2016 and entered the summer with a four-to-one lead in cash on hand, an important advantage in an expensive media market. But Mr. Casten has narrowed the gap considerably.
Other organizations’ ratings:
Previous election results:
| 2016 President | +7 Clinton |
|---|---|
| 2012 President | +8 Romney |
| 2016 House | +18 Rep. |
How our poll result changed
As we reach more people, our poll will become more stable and the margin of sampling error will shrink. The changes in the timeline below reflect that sampling error, not real changes in the race.
One reason we’re doing these surveys live is so you can see the uncertainty for yourself.
If sampling error were the only type of error in a poll, we would expect candidates who trail by one point in a poll of 497 people to win about two out of every five races. But this probably understates the total error by a factor of two.
Our turnout model
There’s a big question on top of the standard margin of error in a poll: Who is going to vote? It’s a particularly challenging question this year, since special elections have shown Democrats voting in large numbers.
To estimate the likely electorate, we combine what people say about how likely they are to vote with information about how often they have voted in the past. In previous races, this approach has been more accurate than simply taking people at their word. But there are many other ways to do it.
Assumptions about who is going to vote may be particularly important in this race.
Our poll under different turnout scenarios
| Who will vote? | Est. turnout | Our poll result |
|---|---|---|
| The types of people who voted in 2014 | 248k | Roskam +6 |
| People whose voting history suggests they will vote, regardless of what they say | 300k | Casten +1 |
| Our estimate | 301k | Casten +1 |
| People who say they are almost certain to vote, and no one else | 311k | Casten +11 |
| People who say they will vote, adjusted for past levels of truthfulness | 327k | Casten +4 |
| The types of people who voted in 2016 | 341k | Casten +3 |
| Every active registered voter | 479k | Casten +5 |
All estimates based on 497 interviews
The types of people we reached
Even if we got turnout exactly right, the margin of error wouldn’t capture all of the error in a poll. The simplest version assumes we have a perfect random sample of the voting population. We do not.
People who respond to surveys are almost always too old, too white, too educated and too politically engaged to accurately represent everyone.
How successful we were in reaching different kinds of voters
| Called | Inter-viewed | Successrate | Ourresponses | Goal | |
|---|---|---|---|---|---|
| 18 to 29 | 2567 | 66 | 1 in 39 | 13% | 10% |
| 30 to 64 | 19093 | 317 | 1 in 60 | 64% | 62% |
| 65 and older | 6675 | 112 | 1 in 60 | 23% | 28% |
| Male | 12210 | 243 | 1 in 50 | 49% | 48% |
| Female | 16170 | 254 | 1 in 64 | 51% | 52% |
| White | 20943 | 359 | 1 in 58 | 72% | 75% |
| Nonwhite | 3891 | 65 | 1 in 60 | 13% | 13% |
| Cell | 20462 | 351 | 1 in 58 | 71% | — |
| Landline | 7918 | 146 | 1 in 54 | 29% | — |
Based on administrative records. Some characteristics are missing or incorrect. Many voters are called multiple times.
Pollsters compensate by giving more weight to respondents from under-represented groups.
Here, we’re weighting by age, primary vote, gender, likelihood of voting, race and region, mainly using data from voting records files compiled by L2, a nonpartisan voter file vendor.
But weighting works only if you weight by the right categories and you know what the composition of the electorate will be. In 2016, many pollsters didn’t weight by education and overestimated Hillary Clinton’s standing as a result.
Here are other common ways to weight a poll:
Our poll under different weighting schemes
| Our poll result | |
|---|---|
| Don’t weight by education, like many polls in 2016 | Casten +3 |
| Our estimate | Casten +1 |
| Don’t weight by primary vote, like most public polls | Even |
| Weight using census data instead of voting records, like most public polls | Even |
All estimates based on 497 interviews
Undecided voters
About 10 percent of voters said that they were undecided or refused to tell us whom they would vote for.
Issues and other questions
Do you approve or disapprove of the job Donald Trump is doing as president?
| Approve | Disapp. | Don’t know | |
|---|---|---|---|
| Voters n = 497 | 41% | 53% | 6% |
Would you prefer Republicans to retain control of the House of Representatives or would you prefer Democrats to take control?
| Reps. keep House | Dems. take House | Don’t know | |
|---|---|---|---|
| Voters n = 497 | 44% | 48% | 7% |
Percentages are weighted to resemble likely voters.
What different types of voters said
Voters nationwide are deeply divided along demographic lines. Our poll suggests divisions too. But don’t overinterpret these tables. Results among subgroups may not be representative or reliable. Be especially careful with groups with fewer than 100 respondents, shown here in stripes.
Gender
| Dem. | Rep. | Und. | |
|---|---|---|---|
| Female n = 254 / 52% of voters | 48% | 40% | 12% |
| Male 243 / 48% | 44% | 49% | 8% |
Age
| Dem. | Rep. | Und. | |
|---|---|---|---|
| 18 to 29 n = 69 / 10% of voters | 64% | 27% | 10% |
| 30 to 44 100 / 17% | 59% | 29% | 11% |
| 45 to 64 218 / 45% | 39% | 51% | 10% |
| 65 and older 110 / 28% | 41% | 50% | 9% |
Race
| Dem. | Rep. | Und. | |
|---|---|---|---|
| White n = 402 / 82% of voters | 46% | 47% | 7% |
| Nonwhite 71 / 13% | 57% | 27% | 16% |
Race and education
| Dem. | Rep. | Und. | |
|---|---|---|---|
| Nonwhite n = 71 / 13% of voters | 57% | 27% | 16% |
| White, college grad 265 / 50% | 48% | 44% | 8% |
| White, not college grad 137 / 32% | 42% | 51% | 7% |
Education
| Dem. | Rep. | Und. | |
|---|---|---|---|
| H.S. Grad. or Less n = 40 / 10% of voters | 55% | 35% | 10% |
| Some College Educ. 126 / 29% | 39% | 53% | 9% |
| 4-year College Grad. 173 / 37% | 43% | 48% | 9% |
| Post-grad. 147 / 23% | 57% | 34% | 10% |
Party
| Dem. | Rep. | Und. | |
|---|---|---|---|
| Democrat n = 137 / 26% of voters | 93% | 4% | 3% |
| Republican 146 / 31% | 5% | 90% | 4% |
| Independent 189 / 38% | 50% | 38% | 12% |
| Another party 12 / 2% | 32% | 29% | 39% |
Primary vote
| Dem. | Rep. | Und. | |
|---|---|---|---|
| Democratic n = 175 / 34% of voters | 83% | 9% | 8% |
| Republican 197 / 41% | 11% | 82% | 7% |
| Other 125 / 25% | 51% | 31% | 18% |
Intention of voting
| Dem. | Rep. | Und. | |
|---|---|---|---|
| Already voted n = 69 / 16% of voters | 54% | 41% | 5% |
| Almost certain 273 / 58% | 49% | 43% | 7% |
| Very likely 103 / 21% | 35% | 53% | 12% |
| Somewhat likely 15 / 2% | 14% | 37% | 49% |
| Not very likely 12 / 1% | 31% | — | 69% |
| Not at all likely 22 / 2% | 38% | 35% | 27% |
Percentages are weighted to resemble likely voters; the number of respondents in each subgroup is unweighted. Undecided voters includes those who refused to answer.
Other districts where we’ve completed polls
About this poll
- Most responses shown here are delayed about 30 minutes. Some are delayed longer for technical reasons.
- The design effect of this poll is 1.14. That’s a measure of how much weighting we are doing to make our respondents resemble all voters.
- Read more about the methodology for this poll.
- Download the microdata behind this poll.
This survey was conducted by The New York Times Upshot and Siena College.
Data collection by Reconnaissance Market Research, M. Davis and Company, the Institute for Policy and Opinion Research at Roanoke College, the Survey Research Center at the University of Waterloo, the University of North Florida and the Siena College Research Institute.
By Michael Andre, Larry Buchanan, Matthew Bloch, Jeremy Bowers, Nate Cohn, Alastair Coote, Annie Daniel, Richard Harris, Josh Katz, Rebecca Lieberman, Blacki Migliozzi, Paul Murray, Adam Pearce, Kevin Quealy, Eden Weingart and Isaac White