[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: dummy coding of categorical variables
From: |
tim.goodspeed |
Subject: |
RE: dummy coding of categorical variables |
Date: |
Thu, 21 Dec 2023 10:18:04 -0000 |
Thank you everyone for your responses.
To close this one off: IT WAS A CODING ERROR. IT IS NOT A BUG.
Took a while last night to find why this was happening. I tried to repeat a
smaller model in excel with the troublesome variables and this resulted in some
similar errors. So it is not a problem with PSPP and was obviously the data.
Thank you all again.
Happy Christmas
Tim Goodspeed
-----Original Message-----
From: EKreyken <ekreyken@gmail.com>
Sent: Wednesday, December 20, 2023 6:44 PM
To: tim.goodspeed@btinternet.com
Cc: pspp-users@gnu.org
Subject: Re: dummy coding of categorical variables results in zero coefficients
and standard errors
Hi Time,
Answering the stats stuff only:
Without seeing the data, I can’t make a definitive call on this, but it seems
that calculations for the t test cannot be completed properly. This indicates
either a) no shared linear variability in your XY variables (incredibly rare to
get perfect 0), forcing the t test to do a division of 0 over the variability
in X, or b) your X variable has no variability (also incredibly rare), forcing
the t test to divide shared linear variability over 0.
You may have one of several problems.
1) dummy coding wasn’t done accurately and you don’t have enough variability in
your X variable
2) you have a perfect non-relationship in those variables (very very rare)
3) you have a non-linear relationship that is really acting weird with the
t-test, but it wouldn’t yield a n/a.
Check the following:
A) get the averages and standard deviations for all predictors. They should
have a standard deviation above 0 for the test to work.
B) plot the data for those specific predictors (X) against the response (Y)
variable. You should end up with 2 separate graphs. Check for linear /non
linear patterns.
C) plot the residuals (google/youtube knows how to do this). Check for
“randomness” vs patterns: patterns in the residuals are bad.
If you’re interested, I offer stats consulting.
Cheers,
Elisabeth
On Dec 20, 2023, at 1:47 AM, tim.goodspeed@btinternet.com wrote:
found?