Analyzing New Poll Numbers from Datanálisis

David Smilde

Barclay’s Research released poll numbers from last week, the first numbers to come out since Chávez’s passing. They show Maduro getting only a limited 2.8 point bump. Datanálisis’s numbers from February had Maduro leading Capriles 46.4% to 34.3% (12.1% difference). These new numbers show Maduro leading Capriles 49.2% to 41.4% (14.4% difference).

However, there are two things to keep in mind about these numbers that make it look a little bit worse for Capriles. First, these numbers include 15-20% of respondents that do not answer. If we assume that those who did not answer will either not vote or break for the two candidates in the same proportion as the existing tendency, then we need to polarize the percentages so that they sum 100%. Doing so we see that in an actual election, the February numbers would have Maduro leading Capriles 57.5% to 42.5% (15% difference), while the March numbers would have Maduro leading Capriles 58.6% to 41.4% (17.2% difference).

Second, the March numbers come from a telephone poll while the February numbers come from Datanálisis’s traditional door-to-door fieldwork. Telephone polls in Venezuela not only have a higher margin or error, they tend to have a class bias. The higher you are in terms of social class, the more likely you are to have a telephone number registered in the type of databases that pollsters use. There are ways of controlling for this. But unless we know if and which sampling controls were used it is hard to know what to do with this comparison. It is always possible that Datanalisis  adjusted correctly or even over-corrected. But past performance of telephone polls in Venezuela suggests it is more likely that the numbers are under-corrected and favor Capriles. These controls are as much art as science.