2016’s state polling ‘his­tor­i­cally bad,’ re­port says

The Washington Post Sunday - - NEWS - BY DAN BALZ dan.balz@wash­post.com

In the wake of last year’s pres­i­den­tial elec­tion, the polling and po­lit­i­cal pro­jec­tion in­dus­tries came in for a beat­ing, ac­cused of miss­ing one of the big­gest po­lit­i­cal sto­ries in a gen­er­a­tion. A new re­port ex­am­in­ing what hap­pened comes to a split con­clu­sion: Na­tional sur­veys were gen­er­ally ac­cu­rate in pro­ject­ing the pop­u­lar vote, but state polls had “a his­tor­i­cally bad year” in fore­cast­ing the re­sults in the elec­toral col­lege.

Don­ald Trump won the pres­i­dency by rolling up 306 elec­toral votes to Hil­lary Clin­ton’s 232. But Clin­ton won the pop­u­lar vote by al­most 3 mil­lion votes and a mar­gin of 2.1 per­cent­age points. This was the sec­ond time in the past five elec­tions when a Demo­crat won the pop­u­lar vote and a Repub­li­can won the elec­toral col­lege.

The re­sults pro­duced a del­uge of crit­i­cism about cam­paign cov­er­age, given near-univer­sal pre­dic­tions of a Clin­ton vic­tory, and why Trump was so un­der­es­ti­mated head­ing up to Elec­tion Day. The au­thors of the re­port summed up the pub­lic re­sponse to the outcome this way:

“The day after the elec­tion, there was a pal­pa­ble mix of sur­prise and out­rage di­rected to­ward the polling com­mu­nity, as many felt that the in­dus­try had se­ri­ously mis­led the coun­try about who would win.”

The crit­i­cism prompted the Amer­i­can As­so­ci­a­tion for Pub­lic Opin­ion Re­search (AAPOR) to con­vene a 13-mem­ber com­mit­tee to study in de­tail to what ex­tent the polls were off and why. The panel’s re­port was re­leased Thurs­day.

On Tues­day, Clin­ton blamed her loss in part on the de­ci­sion by FBI Di­rec­tor James B. Comey to send a let­ter to Congress on Oct. 28 an­nounc­ing that in­ves­ti­ga­tors had come across a fresh trove of her emails and that the in­ves­ti­ga­tion of her use of a pri­vate email server as Sec­re­tary of State was be­ing re­opened.

“If the elec­tion had been on Oc­to­ber 27, I would be your president,” Clin­ton said in a pub­lic in­ter­view with CNN’s Chris­tiane Aman­pour.

The AAPOR re­port sought to an­swer the ques­tion of how much Comey’s let­ter con­trib­uted to Trump’s vic­tory, but it stopped short of pro­vid­ing a con­clu­sive an­swer. “The ev­i­dence for a mean­ing­ful ef­fect on the elec­tion from the FBI let­ter is mixed at best,” the re­port states.

On the eve of the elec­tion, var­i­ous in­di­ca­tors pointed to a Clin­ton vic­tory, ac­cord­ing to the re­port. Among them were the pat­terns of early vot­ing in swing states, fore­cast­ers’ pre­dic­tions and polling data from the Up­per Mid­west, which “showed Clin­ton lead­ing, if nar­rowly, in Penn­syl­va­nia, Michi­gan and Wis­con­sin.” Ul­ti­mately, Trump car­ried those three states by just 77,744 votes, or barely one-half of one per­cent­age point.

The re­port’s two main find­ings about polling un­der­score that it was a treach­er­ous year for poll­sters and there­fore pre­dic­tions. The AAPOR com­mit­tee con­cludes that the na­tional polls “were gen­er­ally cor­rect and ac­cu­rate by his­tor­i­cal stan­dards” and that they were more ac­cu­rate than in 2012. The polls, on av­er­age, pointed to a Clin­ton vic­tory in the pop­u­lar vote by about three per­cent­age points. Her even­tual ad­van­tage was well within the mar­gin of er­ror of the na­tional polls.

But where elec­tions are de­cided, in state-by-state con­tests, things were not so rosy for poll­sters. State polls were his­tor­i­cally bad — the re­port calls it the largest er­ror in state polling of elec­tions start­ing in 2000 — and the key fail­ure was the un­der­es­ti­ma­tion of Trump’s sup­port. This was par­tic­u­larly true in the Up­per Mid­west, where the elec­tion was de­cided.

The AAPOR team also found flaws among those or­ga­ni­za­tions that pro­duce poll ag­gre­gates and pro­jec­tions of re­sults and pointed to th­ese pre­dic­tions as one rea­son so many peo­ple were sur­prised by the outcome of the elec­tion. “They helped crys­tal­lize the er­ro­neous be­lief that Clin­ton was a shoo-in for president,” the re­port states.

The re­port sought to draw a dis­tinc­tion be­tween polls, which the au­thors call “a snap­shot in time,” and pre­dic­tion mod­els. But there was an ex­plicit warn­ing about the lim­its of all such mea­sure­ments. “Cau­tion and hu­mil­ity seem to be in or­der for poll­sters and those who use polls,” the re­port states.

The au­thors sought to determine why Trump’s vote was un­der­es­ti­mated, par­tic­u­larly in state sur­veys. One rea­son, they con­cluded, was that many vot­ers waited un­til the end to de­cide how to vote or changed their minds in the fi­nal days. Th­ese er­rors were “sub­stan­tial and prob­lem­atic in sev­eral con­se­quen­tial states,” the re­port con­cludes.

Among the states where it mat­tered most was Wis­con­sin, a state that Repub­li­cans had not won since 1984 but that Trump car­ried by less than a per­cent­age point. But there were also prob­lems in sev­eral other states that proved cru­cial, in­clud­ing Michi­gan and Penn­syl­va­nia, two other states that had voted for Demo­cratic nom­i­nees in six con­sec­u­tive elec­tions.

Exit polls found that late-de­cid­ing vot­ers backed Trump by sub­stan­tial mar­gins in Florida, Michi­gan, Penn­syl­va­nia and Wis­con­sin in what broke the back of the Clin­ton cam­paign’s path to vic­tory. The re­port notes: “This can be seen as good news for the polling in­dus­try. It sug­gests that many polls were prob­a­bly fairly ac­cu­rate at the time they were con­ducted.”

That might be small con­so­la­tion to the polling com­mu­nity, given pub­lic per­cep­tions that fore­casts in gen­eral were off the mark in 2016.

One clear weak­ness in state polls, ac­cord­ing to the re­port, was the fail­ure of many poll­sters to weight their data on the ba­sis of ed­u­ca­tional achieve­ment. In 2016, there was a sig­nif­i­cant ed­u­ca­tion gap in vot­ers’ be­hav­ior, with non-col­lege-ed­u­cated vot­ers sup­port­ing Trump heav­ily in com­par­i­son to those with col­lege de­grees.

Many state polls con­tin­ued to base their re­sults on sam­ples that in­cluded too many col­lege-ed­u­cated vot­ers and too few with­out de­grees, de­spite the fact that there was a well-doc­u­mented ed­u­ca­tion gap early in the cam­paign.

An­other pos­si­ble cause for er­rors in state polls was the change in the com­po­si­tion of the elec­torate and of turnout by dif­fer­ent groups of vot­ers, com­pared with 2012. The re­port sug­gests that this was a prob­lem, but the au­thors said they lacked some ad­di­tional data from the fed­eral govern­ment that would al­low for more de­tailed study.

A much smaller fac­tor cited as a pos­si­ble rea­son polls in the Mid­west tended to un­der­es­ti­mate sup­port for Trump was that he ap­peared as the first name on bal­lots in Florida, Michi­gan and Wis­con­sin.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.