Posted by
GoffResearch on Friday, October 20, 2006 5:55:46 PM
I am working on a theory. It is just
a tentative theory at this point and hard to evaluate. However, I think it has
merit and is applicable to the current situation.
When Republicans face a hostile
media environment (like now and during Vietnam) a small percentage of
conservatives and also Republicans will refuse to answer polls. As a result of
replacement sampling by pollsters, a higher percentage of people actually polled
will self identify as Democrats. Current telephone polls rarely get results from
more than 20% of the original random sample. That means that up to 80% of the
sampled voters are not (exactly) randomly selected, but exhibit some response
bias.
There may be some real shifts in
party identification. However, in a hostile media environment there may also be
a cross current that results from pollsters gathering data from a slightly less
conservative sample than the population who actually vote in the elections. In
other words, due to replacement sampling (the replacements are less conservative
and less Republican than the actual voting population) the whole sample shifts
to the left slightly. In a hostile media environment pollsters will overstate
Democratic strength in most (not all) Congressional
Districts.
There are several anecdotal reasons
to believe such a shift may occur.
1. In hostile media environments the
left tend to get vocal and some of the right tend to become silent (i.e., silent
majority). 2. Political orientation
(conservative etc.) tends to be stable over time. 3. True party identification tends
to be fairly stable.
4. Pew and some other pollsters
measure party identification as a current attitude not as a behavior. That makes
party affiliation appear to move a lot
more than it actually does (as measured by actual voting behavior).
5. They also only measure three
categories: Democrat, Republican and Independent. That creates more measurement
error as Greens, Libertarians, and other
party affiliations may not identify with the three categories.
The results are polls that show a
slight bias for Democrats especially in a hostile media environment. Republican
voting strength could be understated by 1 to 3 points. After an election occurs
and Republicans do better than expected, this bias is usually ignored or
attributed to superior Republican turnout. I would suggest it is most likely to
occur at the CD level in off year elections. Partly because accurate polling at
CD level in off years is very difficult. The task of assessing who is likely
to vote is more problematic in the CD races in off years. I also suspect
that the response bias for most CD polls is greater than for Senate or
Presidential polls as the actual under vote in most districts tends to be higher.
Use this if it is helpful. I am not
sure if it can be tested. It is a question of measurement error in polling vs
turnout.
However on balance Republicans may
be doing slightly better than currently thought.
Brent Goff