Raising questions

Interesting figures come out today in the two Brazilian papers I usually check, the Jornal do Brasil and the Folha de Sao Paulo. Both have prompted me to write to the polling firms involved to see if I can establish why there may be differences in the results.

In the JB the present mayor, the PFL's Cesar Maia, is reported to have fallen 11 points to 34% between May and June according to a poll done by Instituto Gerp. The second placed candidate, the PL's Marcelo Crivella, rose 3 points, to 14%. And in third place, former mayor, Luiz Paulo Conde of the PMDB has fallen two points to 6%.

Over in the Folha, both front-runners appear to be doing better: Cesar Maia is given 38% on Datafolha's most recent poll (which doesn't appear to be on its website), while Marcelo Crivella is on 20% and Conde on 9%. Even allowing for a margin of error of 3.1%, that's quite a difference.

Meanwhile in the Sao Paulo race Datafolha's most recent figures gave Jose Serra of the PSDB 30%, Paulo Maluf 24% and the PT's Marta Suplicy 20%, according to a poll taken last Friday and Saturday.

But Ibope's figures are slightly different. While they still give Serra 30% and Maluf isn't far out with 21%, Marta is presented at being on 16%. Again the claim is made that there is a margin of error of 3% - but how can Marta's figures not hold up compared to the other two?

Of the three studies, I've only found Ibope's. Maybe the clue lies in the timing of their poll, which took place a week earlier than Datafolha's. But could that account for the slump in Marta's support?

Ibope also states that the process of deciding who to interview falls in two stages: first, they established the proportion of the area to be surveyed (Sao Paulo) according to census data. They then made a sample of 1024, which reflected that data. The criteria they used included the following: gender, age range, educational level, type of employment and geographic location. They claim that their margin of error for this poll was lower than that of Datafolha's or Instituto Gerp's - at 2.8 points.

However, other than stating they did these interviews personally, they don't specify how it was done. Was it face-to-face? Or by telephone? And how many did they have to rule out, who refused to take part?

Of course, these are questions I'm sure others would like to know about the other polls too. While Ibope has at least given some details of its sampling, it would be helpful to know whether the sample has changed and whether it takes account of those likely to vote (an ongoing challenge facing pollsters in the UK).