>Doug:
>> average 47 45 4 2
>>
>> actual 48 48 2 0 0
>
>I assume that these are unweighted averages. I wonder, however, what these
>averages would be if they were weighted by the sample size. Such weighting
>would have the practical effect of increasing the sample size to that of the
>total number of people polled in all surveys. It may (or may not) correct
>for a possible bias in individual surveys.
There's probably not that much difference in sample size. Regular polls are usually around 1000, and tracking polls are typically three-day averages of smaller samples (though Zogby's tracking poll uses a sample of 1200).
To answer Charles B's earlier question, as far as I know, pollsters didn't change their techniques after 2000. The errors were in line with earlier years, and a 2-point miss is really pretty small.
Doug