Ainda sobre sondagens e resultados eleitorais

Ver aqui três análises (.pdf) das sondagens nas eleições do Reino Unido onde não se dispensa uma comparação com os resultados das eleições, e onde se usa essa comparação para reflectir do ponto de vista metodológico.

Algumas lições para nós:

1. A informação pública e divulgada acerca das metodologias utilizadas no Reino Unido é incomensuravelmente mais completa do que aquela que é divulgada em Portugal;

2. O facto de se reconhecer que sondagens são "fotografias" de um momento e não "previsões" não impede que a comparação com resultados eleitorais seja informativa do ponto de vista metodológico;

3. Mas isso não impede por sua vez, a tensão entre a busca de "boas fotografias" ou "boas previsões" nas diferentes práticas dos institutos de sondagens, gerando confusão entre a opinião pública e diminuindo a utilidade da análise das sondagens à luz dos resultados eleitorais. Sobre isto, Nick Sparrow, da ICM:

"When pollsters come up with a final poll that is close to the outcome of an election, with much pride they say their predictions were accurate. When they get it wrong they may often say that the poll was right when it was taken but was a snapshot of a moving situation. Clients who have sponsored an accurate poll will also use the opportunity to poke fun at their competitors who may have carried a somewhat less accurate prediction.

Some pollsters argue that all polls are snapshots but their final poll is a prediction. If this leads to rather different methods being employed on the final poll, it would seem to undermine any claim to accuracy of any poll except one taken immediately before an election.
(...)

If polls are snapshots that report what people think they would do in an immediate election, the fact that they have produced estimates a little too kind to Labour over the last few years seems more defendable, simply on the basis that they measure support for each party at any given point in time, and cannot accurately predict who will state that support in the polling booth. If they claim to be predictions then the persistent Labour bias needs to be addressed. The pollsters cannot idly flit between the two.

One solution would be for pollsters and their clients to publish snapshots AND predictions. It would seem to be quite fair to point out to readers where the uncertainties lie in the data and what direction of error those uncertainties imply.

(...)
Perhaps pollsters and their clients should openly do both, telling consumers of polls what the data says, but also adding notes of caution about any data that may lead to a somewhat different prediction of the actual outcome.

4. Talvez se em Portugal houvesse uma instituição de natureza semelhante ao British Polling Council no Reino Unido ou ao National Council on Public Polls americano muitos destes assuntos e problemas pudessem ser resolvidos de forma mais transparente para a opinião pública...