Friday, June 8, 2007

How Well Did The Public Opinon Pollsters Do in 2007 Elections?



Although the counting is not yet over and only ten of the twelve winning senators have been proclaimed, I think it is possible to compare the numbers emerging from the elections and the predictions of the survey firms. In the plot above, you can see a graphical representation of three data sets:

(1) the PDI/SWS May 1-2 pre-election poll.
(2) the Namfrel Tally with 87.69% of 224,748 precincts already tallied.
(3) the ABSCBN/Pulse Asia Exit Poll conducted after the voting on May 14.

Below is the numerical data from which the graph above was plotted.
SENATORIAL CANDIDATE SWS/PDI May Poll NAMFREL Batch 43 ABSCBN/PULSE Exit Poll

Legarda

58.0 58.5 58.5
Escudero 43.0 57.4 53.3
Lacson 39.0 49.6 46.4
Villar 46.0 48.3

49.8

Aquino 36.0 46.0 42.6
Pangilinan 41.0 45.9 44.6
Angara 31.0 40.4 41.1
Cayetano 34.0

37.3

31.0
Honasan 32.0 37.1 34.6
Arroyo 31.0 37.0 36.8
Trillanes 24.0 35.6 35.4
Pimentel 25.0 34.6 28.5
Zubiri 32.0 33.9 34.9

Recto

36.0 33.1 34.3
Defensor 25.0 31.1 28.2
Pichay 23.0 30.6 30.4
Roco 23.0 27.0 28.4

Both SWS and Pulse predicted a 6-4-2 GO-TU-IND outcome. But that has not come to pass and it looks most likely that the result will be 8-2-2 GO-TU-IND. The exit poll, as expected was much more accurate in predicting outcomes for individual senators, though it's predicted composition also did not come to pass.

A note on how I did this even without knowing the actua totall turnout of voters: I assumed that the total of Loren Legarda in the Namfrel tally, some 15,200,000 votes corresponds to the 58.5% both surveys were predicting she would get, then computed the percentages implied for each of the other candidates based on their actual votes, to see the pattern predicted by each.

The margin of error for the SWS survey was plus or minus 3 percent. For the ABSCBN/Pulse Exit Poll, the margin of error was reportedly plus or minus 2 percent.

A more detailed comparative analysis can be made once the final numbers and winners are in, but at least one interesting general observation can be made:

Look at the Namfrel Curve of the winning senators. It is MONOTONICALLY DECREASING from left to right, because of course, the Namfrel count is in the correct order from highest percentage garnered (Loren Legarda) to the lowest (at No. 17) in this plot. But you can easily spot where the May survey and Exit poll are wrong, because neither one has the correct order or composition, with the SWS/PDI pre-election survey missing for example the winning candidacy of Antonio Trillanes and the defeat of Ralph Recto.

No comments: