Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

shifting categories over time

0 views
Skip to first unread message

Amy

unread,
Aug 31, 2010, 6:23:56 PM8/31/10
to
I have 2 years worth of survey measures on a fair number of variables
(75-ish). Data re categorized by "type". (Different respondents,
same questions.)

In year 2008 there were 16 types. In 2010, there were 32 types,
several of which are re-combinations of the 2008 types. I need to
compute a variable "type" that categorizes all the types, where, for
example 2010's TYPE Z is the average of the scores for 2008 TYPE B and
TYPE C.

I want 1 variable that allows me to compare means over time, where
type Z's 2010 data is matched to the average of 2008's average of TYPE
B and C.

Many thanks for your help in advance,

Amy

Ray Koopman

unread,
Aug 31, 2010, 7:17:23 PM8/31/10
to

You're setting yourself up for Simpsons's paradox if the relative
number of B's & C's you would have observed in 2010 using the 2008
classification rules differs from the relative weight assigned to the
2008 B & C means when you average them. In general, the only types
that can be compared are those for which the classification rules have
not changed.

Amy

unread,
Sep 1, 2010, 8:04:17 AM9/1/10
to


Yes, this makes sense. I had to google Simpson's Paradox, but
remembered it once I did. Many thanks!!

Amy

0 new messages