terça-feira, outubro 21, 2008

Podemos acreditar no que estamos a ver? Depende.

Os textos no post abaixo sobre as possibilidades de recuperação de McCain pressupõem que aquilo que as sondagens nos estão a dizer é, "na média", verdade. Que Obama tem uma vantagem de mais de 5 pontos sobre McCain, que Obama subiu drasticamente desde a crise financeira, que McCain vem recuperando nos últimos dias.

Mas se isto tudo, ou parte disto, for mentira? E se houver um enviesamento sistemático nos resultados? Uma das fontes de enviesamento já aqui muito discutida é o chamado Bradley Effect, através do qual a vantagem de Obama sobre McCain poderá estar a ser sobrestimada. A segunda é o recurso a telefones fixos, através do qual a vantagem de Obama sobre McCain poderá estar a ser sistematicamente subestimada.

James Stimson, da UNC em Chapel Hill, recorda-nos um terceiro possível enviesamento, que desta vez tem repercussões naquilo que julgamos estar a observar nos últimos dias: as pressuposições sobre os votantes prováveis que estão inseridas nos modelos.

The entry of new tracking polls (with likely voter estimates) and the changeover of existing ones to likely voter estimates are confusing the situation this week. Because Republicans traditionally turn out at higher rates than do Democrats, the likely voter estimates shift the composition of samples in the Republican direction, producing better numbers for McCain. This is creating what appears to be an illusory trend toward McCain support, an apparent decline in the Obama lead. This is seen most clearly in the Gallup organization's now triple daily reports. With a pleasing transparency, Gallup is reporting registered voters numbers (as before) and two different likely voter estimates. One Gallup calls the traditional filter assumes that historic turnout patterns will prevail and thus African-Americans and young people, for example, are assumed to turn out at relatively low rates (compared to the perennial turnout champs, the middle-aged and middle class.) Knowing that such assumptions are likely to be inaccurate this year, Gallup is also using a filter (called Expanded) that is based only on what respondents say about voting intentions. I commend Gallup for leaving us free to make our own assumptions about turnout, rather than imposing its own on us (and usually without any information about what they are.) From its latest release (10/19) Gallup shows an Obama lead of 10 points (52-42) among registered voters, 7 points (51-44) among the Expanded likely voter set and only 3 points (49-46) with the Traditional likely voter filter. It makes quite a difference which one gets reported. There is no trend in the registered voter estimates. But if you compare the older registered voter numbers with the newer likely voter numbers the illusion of trend appears.
Enviar um comentário