Im Jahre 2004 kamen die Umfrageserien, die bis vor den Wahltag erstellt worden waren, im Schnitt bis auf 2 Prozentpunkte an das effektive Ergebnis heran. Sie waren damit im Schnitt etwas besser als vier Jahre zuvor. In der Richtung haben die Umfrageserien von 2000 und 2004 jeweils den Republikaner Bush leicht überschätzt. Auch das spricht für eine Wahlsieg von Barack Obama, der in den letzten Umfragen mit durchschnittlich 7,6 Prozent führt.
6 der 8 Institute, die 2004 eine Projektion erstellten, sahen richtigerweise Georges W. Bush (50.7%) als Sieger vor John Kerry (48.3%). Am genauesten waren damals die Vorhersage von TIPP (50.1 zu 48.0). Sie gab Bush 2.1 Prozentpunkte Vorsprung. Mit etwas abnehmender Genauigkeit folgten damals die Institute PEW, Battleground-Tarrence und Harris, beide knapp vor Zogby und Gallup. Eigentliche Fehlprognosen lagen bei Democracy Corps und Battleground-Lake vor.
Im Jahr 2000 war die Sache komplizierter, weil Al Gore (48.4%) effektiv einen halbe Prozentpunkt mehr Wählerstimmen hatte als Georges W. Bush (47.9%). Diese wurde dank eine hauchdünnen Mehrheit bei den Elektoren gewählt. Die Umfragen wiederum sahen Bush ist als klaren Sieger. Nur Zogby hatte Kerry vorne, und Harris kam dem bizzaren Endresultat mit 47:47 am genauesten.
Was lernt man daraus?
Erstens, die Differenz zwischen den beiden Spitzenkandidaten wurde 2000 falsch, 2004 aber richtig erkannt. Der Fehler liegt zwischen 2 und 3 Prozentpunkten.
Zweitens, die republikanischen Bewerber werden nicht einfach unterschätzt, egal ob sie Herausforderer oder Amtsinhaber sind.
Drittens, auf ein Institut abzustellen, ist nicht einfach, da Harris nicht mehr dabei ist, und TIPP und TIPP“>Zogby, die beiden besten bei einer Wahl bei der anderen kleinere Probleme hatten.
Wenn Obama diesmal in allen Umfrageserien mit durchschnittlich 7,6 Prozent (wenn auch mit unterschiedlichen Differenzen von 2 bei Battleground Tarrence bis 11 Prozent bei Zogby resp. Gallup führt, kann, egal wie gross der Vorsprung letzten Endes sein wird, nichts mehr schief gehen.
Weder für ihn, noch für die Umfrageinstitute als Ganzes.
Claude Longchamp
Was halten Sie denn von diesem Ich-hab-peinlicherweise-vergessen-wie-er-heisst-Effekt, der besagt, dass bei nichtweissen Kandidaten durchaus ein weitaus grösserer «Gap» zwischen Umfragewerten und konkreten Resultaten entstehen kann? So aus dieser «doch doch, ich wähl den schon, ich bin doch tolerant und offen»-Haltung raus, die dann eben doch nicht immer stimmt.
Bradley-Effekt meinen Sie sicher.
american wikipedia fasst das Problem wie folgt zusammen:
“The causes of the polling errors are debated, but pollsters generally believe that perceived societal pressures have led some white voters to be less than forthcoming in their poll responses. These voters supposedly have harbored a concern that declaring their support for a white candidate over a non-white candidate will create a perception that the voter is racially prejudiced. During the 1988 Jackson presidential campaign, Murray Edelman, a veteran election poll analyst for news organizations and a former president of the American Association for Public Opinion Research, found the race of the pollster conducting the interview to be a factor in the discrepancy. Edelman’s research showed white voters to be more likely to indicate support for Jackson when asked by a black interviewer than when asked by a white interviewer.
Andrew Kohut, who was the president of the Gallup Organization during the 1989 Dinkins/Giuliani race and later president of the Pew Research Center, which conducted research into the phenomenon, has suggested that the discrepancies may arise, not from white participants giving false answers, but rather from white voters who have negative opinions of blacks being less likely to participate in polling at all than white voters who do not share such negative sentiments with regard to blacks.
While there is widespread belief in a racial component as at least a partial explanation for the polling inaccuracies in the elections in question, it is not universally accepted that this is the primary factor. Peter Brodnitz, a pollster and contributor to the The Polling Report newsletter, worked on the 2006 campaign of black U.S. Senate candidate Harold Ford, Jr., and contrary to Edelman’s findings in 1988, Brodnitz indicated that he did not find the race of the interviewer to be a factor in voter responses in pre-election polls. Brodnitz suggested that late-deciding voters tend to have moderate-to-conservative political opinions and that this may account in part for last-minute decision-makers breaking largely away from black candidates, who have generally been more liberal than their white opponents in the elections in question. Another prominent skeptic of the Bradley effect is Gary Langer, who serves as the director of polling for ABC News. Langer has described the Bradley effect as “a theory in search of data.” He has argued that inconsistency of its appearance, particularly in more recent elections, casts doubt upon its validity as a theory.
With regard to the 1982 contest between Bradley and Deukmejian in particular, people involved with both campaigns, as well as those involved with the inaccurate polls have refuted the significance of the Bradley effect in determining that election’s outcome. Mark DiCamillo, Director of the The Field Poll, which was among those that had shown Bradley with a strong lead, has not ruled out the possibility of a Bradley effect as a minor factor, but also said that the organization’s own internal examination after that election identified other possible factors that may have contributed to their error, including a shift in voter preference after the final pre-election polls and a high profile ballot initiative in the same election, a Republican absentee ballot program and a low minority turnout, each of which may have caused pre-election polls to inaccurately predict which respondents were likely voters.
Prominent Republican pollster V. Lance Tarrance, Jr. flatly denies that the Bradley effect occurred during that election, echoing the absentee ballot factor cited by DiCamillo. Tarrance also reports that his own firm’s pre-election polls done for the Deukmejian campaign showed the race as having closed from a wide lead for Bradley one month prior to the election down to a statistical dead heat by the day of the election. While acknowledging that some news sources projected a Bradley victory based upon Field Poll exit polls which were also inaccurate, he counters that at the same time, other news sources were able to correctly predict Deukmejian’s victory by using other exit polls that were more accurate. Tarrance claims that The Field Poll speculated, without supplying supporting data, in offering the Bradley effect theory as an explanation for why its polling had failed, and he attributes the emergence of the Bradley effect theory to media outlets focusing on this, while ignoring that there were other conflicting polls which had been correct all along.
Sal Russo, a consultant for Deukmejian in the race has said that another private pollster working for the campaign, Lawrence Research, also accurately captured the late surge in favor of Deukmejian, polling as late as the night before the election. According to Russo, that firm’s prediction after its final poll was an extremely narrow victory for Deukmejian. He asserts that the failure of pre-election polls such as The Field Poll arose, largely because they stopped polling too soon, and that the failure of the exit polls was due to their inability to account for absentee ballots.
Blair Levin, a staffer on the Bradley campaign in 1982 said that as he reviewed early returns at a Bradley hotel on election night, he saw that Deukmejian would probably win. In those early returns, he had taken particular note of the high number of absentee ballots, as well as a higher-than-expected turnout in California’s Central Valley by conservative voters who had been mobilized to defeat the handgun ballot initiative mentioned by DiCamillo. According to Levin, even as he heard the “victory” celebration going on among Bradley supporters downstairs, those returns had led him to the conclusion that Bradley was likely to lose.
In 2008, several political analysts discussing the Bradley effect referred to a study authored by Daniel J. Hopkins, a post-doctoral fellow in Harvard University’s Department of Government, which sought to determine whether the Bradley effect theory was valid, and whether an analogous phenomenon might be observed in races between a female candidate and a male candidate. Hopkins analyzed data from 133 elections between 1989 and 2006, compared the results of those elections to the corresponding pre-election poll numbers, and considered some of the alternate explanations which have been offered for any discrepancies therein. The study concluded finally that the Bradley effect was a real phenomenon, amounting to a median gap of 3.1 percentage points before 1996, but that it was likely not the sole factor in those discrepancies, and further that it had ceased to manifest itself at all by 1996. The study also suggested a connection between the Bradley effect and the level of racial rhetoric exhibited in the discussion of the political issues of the day. It asserted that the timing of the disappearance of the Bradley effect coincided with that of a decrease in such rhetoric in American politics over such potentially racially-charged issues as crime and welfare. The study found no evidence of a corresponding effect based upon gender – in fact, female Senate candidates received on average 1.2 percentage points more votes than polls had predicted.”
Meine Position ist: Wenn die Umfrageinstitute seither nicht gelernt haben, auf solche Probleme zu reagieren, haben die amerikanischen Kollegen ein Problem.
Herzlichen Dank – genau den meinte ich. Das ist ja beruhigend, wenn man damit auch bei den Umfrageinstituten mittlerweile umzugehen weiss; ich meine mich zu erinnern, dass der Bradley-Effekt während des Obama-Clinton-Rennens in den Primaries wiederholt zitiert wurde (allerdings wohl mehr als Befürchtung denn als tatsächlich eintreffendes Phänomen).
Ja, das stimmt. Er wurde auch in den letzten 10 Tagen von Republikanern eingesetzt. Das macht es verdächtig. Denn es sollte heissen: Obama ist weniger gut als in den Umfragen.