De­spite crit­i­cism, elec­tion polls beat past per­for­mances

The Washington Post - - ELECTION 2018 - BY SCOTT CLEMENT

Tues­day’s vic­to­ries for Democrats in the U.S. House and Repub­li­cans in the Se­nate made for a much less sur­pris­ing elec­tion night than Pres­i­dent Trump’s elec­tion two years ago, but that didn’t pre­vent a re­run of the de­bate over whether poll­sters had missed the trends that be­came clear late Tues­day.

Against re­newed crit­i­cism from some ca­ble com­men­ta­tors, polling ex­perts of­fered at least tepidly pos­i­tive ap­praisals of polling per­for­mance on Wed­nes­day, as bal­lots were still be­ing tal­lied across the coun­try. A Wash­ing­ton Post pre­lim­i­nary anal­y­sis of poll ac­cu­racy across U.S. Se­nate and gu­ber­na­to­rial races sug­gests that er­rors were smaller than for sim­i­lar con­tests in the past, in­clud­ing state pres­i­den­tial polls in 2016, but that these sur­veys did tend to un­der­es­ti­mate Repub­li­can can­di­dates more than Demo­cratic ones this year.

Polls have faced height­ened scru­tiny since the 2016 elec­tion, when they were widely panned as in­ac­cu­rate in fail­ing to pre­dict Trump’s vic­tory. The crit­i­cism was partly mis­placed: Na­tional polls ac­tu­ally proved to be ac­cu­rate in es­ti­mat­ing a small ad­van­tage for Hil­lary Clin­ton, who won the pop­u­lar vote by 2.1 per­cent­age points. But polls in key bat­tle­ground states con­sis­tently un­der­es­ti­mated Trump’s sup­port, pro­vid­ing an un­com­fort­able re­minder that such sur­veys have his­tor­i­cally been less ac­cu­rate.

Early skep­ti­cism played out promi­nently on CNN on Tues­day night. “I’m struck at how wrong polling was in so many places,” said host Jake Tap­per. “Ob­vi­ously there were polls that showed An­drew Gil­lum win­ning the gover­nor’s of­fice in Florida.” Tap­per con­cluded that “once again, it does seem like there are some Repub­li­can votes that peo­ple are not able to pick up. . . . They’re al­ways polling wrong in that di­rec­tion.”

Shortly af­ter­ward, Demo­cratic strate­gist David Ax­el­rod pre­dicted that poll per­for­mance “is go­ing to prompt an­other round of soul-search­ing about whether and how you can poll ac­cu­rately, be­cause a lot of these races that were blowouts tonight or ap­par­ently blowouts tonight polled as tough races.” He noted, how­ever, that the Florida poll re­sults, which Tap­per had crit­i­cized, “ac­tu­ally fell within the mar­gin of er­ror.”

Ax­el­rod of­fered a more pos­i­tive as­sess­ment Wed­nes­day morn­ing, say­ing in an email to The Wash­ing­ton Post, “Within mar­gins of er­ror, I think polls were rea­son­ably good.” CNN de­clined to com­ment Tues­day on Tap­per’s char­ac­ter­i­za­tions of poll ac­cu­racy.

Two polling ex­perts reached by The Post on Tues­day praised the Tues­day show­ing.

“Elec­tion polling had a pretty good night,” said Scott Keeter, a se­nior sur­vey ad­viser at the Pew Re­search Cen­ter. “The big-pic­ture story from the polls turned out to be ac­cu­rate: Polling showed that the Democrats had a very good chance to take the House back and that Repub­li­cans had a very good chance to hold or in­crease their ad­van­tage in the Se­nate. Both were cor­rect.”

Fi­nal pre-elec­tion sur­veys tracked by RealClearPol­i­tics found Democrats led Repub­li­cans by seven per­cent­age points when vot­ers were asked which party’s generic House can­di­date they would sup­port. The lat­est con­gres­sional vote to­tals put Democrats’ ad­van­tage at four points, though this is ex­pected to grow as votes from Cal­i­for­nia and Wash­ing­ton are fully tal­lied. The New York Times fore­casts that once all bal­lots have been counted, Democrats will win the pop­u­lar vote by seven points, iden­ti­cal to the av­er­age mar­gin in pub­lic sur­veys.

Demo­cratic poll­ster An­drew Bau­mann called the pre-elec­tion polls “quite ac­cu­rate, par­tic­u­larly for a midterm that ended up be­ing to­tally dif­fer­ent than any pre­vi­ous midterm.” Bau­mann, who con­ducts sur­veys for the Global Strat­egy Group, pointed to Florida as an ex­cep­tion.

Repub­li­can poll­ster Kris­ten Soltis Anderson was less pos­i­tive, say­ing that she ini­tially gave polls a “C-plus” for their ac­cu­racy but that “a B-mi­nus or B is per­haps more fair.” Anderson, a part­ner at the firm Ech­e­lon In­sights, said that even though pre-elec­tion polls largely met her ex­pec­ta­tions, a hand­ful of high-pro­file sur­prises oc­curred, with elec­tion re­sults di­verg­ing from fi­nal polling av­er­ages, in­clud­ing the Repub­li­can wins in Florida and large Repub­li­can win­ning mar­gins in In­di­ana and Mis­souri.

In Florida’s Se­nate race, Sen. Bill Nel­son (D) held a lead of roughly three points across 17 polls con­ducted since mid-Oc­to­ber. At one point Wed­nes­day af­ter­noon, he was 0.3 points be­hind his Repub­li­can op­po­nent, Gov. Rick Scott, in the vote count. While most sur­veys’ re­sults were within the range of sam­pling er­ror — which is nearly twice the tra­di­tional mar­gin of sam­pling er­ror when com­par­ing both can­di­dates’ sup­port — as a whole, they un­der­es­ti­mated Scott’s stand­ing.

Keeter, of Pew, noted that while er­rors oc­curred in some of the most prom­i­nent Se­nate and gu­ber­na­to­rial races, polls did not al­ways un­der­es­ti­mate Repub­li­cans. He pointed out that al­though polls were a bit too Demo­cratic in Florida, polls in Texas had un­der­stated Demo­crat Beto O’Rourke’s chances in the Se­nate race there.

Fi­nal Texas polls be­fore the elec­tion found the Repub­li­can in­cum­bent, Sen. Ted Cruz, lead­ing by be­tween three and 10 per­cent­age points. O’Rourke lost by less than three per­cent­age points, ac­cord­ing to Wed­nes­day’s vote tal­lies.

The Post’s anal­y­sis of the bat­tle­ground U.S. Se­nate and gover­nor polls tracked by RealClearPol­i­tics finds that on the ba­sis of pre­lim­i­nary vote counts, state polls erred by an av­er­age of 4.1 per­cent­age points in es­ti­mat­ing the vote mar­gin be­tween Demo­cratic and Repub­li­can can­di­dates. That is lower than an av­er­age er­ror of 5.1 points in 2016 state-level pres­i­den­tial polls, but is just about av­er­age for state-level polls in pres­i­den­tial elec­tions since 2000. Polling er­rors in key races for the Se­nate and gov­er­nor­ships were also some­what lower than the av­er­age of 5.4 points for Se­nate and gover­nor elec­tions since 1998, ac­cord­ing an anal­y­sis by the web­site FiveThir­tyEight this spring.

As in 2016, some of the larger polling er­rors this year oc­curred in states where polling was scarce. In Ohio, sur­veys by Emerson Col­lege and Gravis Mar­ket­ing were the only two polls within three weeks of the elec­tion, both con­ducted us­ing less-ex­pan­sive au­to­mated and on­line meth­ods. They found Demo­crat Richard Cor­dray up by three to five per­cent­age points, while on Tues­day Repub­li­can Mike DeWine won by a sim­i­lar mar­gin.

On av­er­age across com­pet­i­tive elec­tions for the U.S. Se­nate and gov­er­nor­ships, polls un­der­es­ti­mated Repub­li­cans’ vote mar­gins in these con­tests by two per­cent­age points, the equiv­a­lent of a poll’s find­ing a Repub­li­can trail­ing 49 per­cent to 51 per­cent when the vote re­sult is an even 50-50.

One bright spot this year ap­pears to have been con­gres­sional district polling, which saw a boon this year de­spite the greater dif­fi­culty of con­duct­ing sur­veys at this nar­row geo­graphic level. Bau­man pointed to the am­bi­tious ef­fort by the New York Times and Siena Col­lege to con­duct more than 90 sur­veys in con­gres­sional bat­tle­ground dis­tricts as a suc­cess, not­ing that they “ended up be­ing quite ac­cu­rate and re­ally did a good job of ed­u­cat­ing the po­lit­i­cal pub­lic about the dif­fi­cul­ties of do­ing polling well.”

Mon­mouth Univer­sity’s in­creased ef­fort to sur­vey con­gres­sional dis­tricts also ap­pears to have paid off, with fi­nal polls prov­ing largely ac­cu­rate or con­sis­tent with the di­rec­tions in which races were trend­ing. Emily Guskin con­trib­uted to this re­port.


An­drew Gil­lum, the Demo­cratic Party nom­i­nee for gover­nor of Florida, leaves the stage Tues­day night at Florida A&M Univer­sity after con­ced­ing the race to Repub­li­can nom­i­nee Ron DeSan­tis.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.