Re: Weighting Oversamples..

mike wrote:
I was hoping someone here could help with the methodology involved with
handling an "oversample" of a dataset.

I am dealing with a political polling public opinion survey where
N=400, thus the poll consisted of Democrats/Republicans and
Independents. There was an oversample of 200 additional Democratic
Women and I have been instructed to "Weight the survey back to N=400 so
that the Democratic women are representative of the total.". In other
words, they want the oversample to be included in the analysis, but
they dont want to see the data "skewed" towards Democratic Women. So,
I was hoping that someone could verify my logic and methodology to do
this.

Here is the procdedure I am using.

1.) Get the distribution of the base sample (N=400) party
identification and gender which is
Rep = 42%
Dem=42%
Indepenedent =16%

Male = 48%
Female = 52%

2.) Get the distribution of the base sample + over sample (N=600)
which is
Rep = 28%
Dem=53%
Independent = 20%

Male = 36%
Female = 64%

3.) Weight the combined data (base sample+oversample, N=600) to be the
same distribution as the base sample. Thus, when I run the weighted
frequencies with the data including the oversample it looks just like
the base sample which was.
Rep = 42%
Dem=42%
Indepenedent =16%

Male = 48%
Female = 52%

Does this sound right? Is it statistically sound? I dont see how else
I could weigh in that oversample data? Any input/advice/verification
would be greatly appreciated!

Thanks in advance.

Mike


Weighting will result in decreased precision by "undoing" your actual sample size. If party and sex and important, why not give party- and sex-specific answers and no do weighting?

Frank Harrell
.

Relevant Pages

  • Re: Weighting Oversamples..
    ... I am dealing with a political polling public opinion survey where ... There was an oversample of 200 additional Democratic ... Women and I have been instructed to "Weight the survey back to N=400 so ... Get the distribution of the base sample + over sample ...
    (sci.stat.math)
  • Re: Weighting Oversamples..
    ... There was an oversample of 200 additional Democratic ... Women and I have been instructed to "Weight the survey back to N=400 so ... Indepenedent =16% ... Get the distribution of the base sample + over sample ...
    (sci.stat.math)
  • Weighting Oversamples..
    ... There was an oversample of 200 additional Democratic ... I was hoping that someone could verify my logic and methodology to do ... Indepenedent =16% ... Get the distribution of the base sample + over sample ...
    (sci.stat.math)