I'm running a complex ordinal regression in SPSS (I know this isnt an ideal programme but needs must). Anyway I am getting output with a huge design effect number. Does anyone know why!? It is literally like 3 million and something.
Are there some units ( eg your Primary Sampling Units) which are completely one category so that there is no variation?
Do a crosstab on the units by the ordinal categories .
Good luck!
More info is needed to answer... What specific test(s) are you conducting? What's the sample size? Did the data meet assumptions? Etc?
Does it look like this?
- 1.189383e-06
Then maybe your output is ridiculously low as it is the scientific notation of
- 0.000001189383
Otherwise I'd say we need more info.
I dont think it is to do with the PSUs either bc I ran a crosstab and although some categories have 0 frequencies there are 8 regions used in sampling and 7 levels to fluid intelligence and it appears there is variation therein.
The sample size is 7331, the DV is fluid intelligence and its not exactly normally distributed but the VIF and tolerance does not indicate multicollinearity. The parallel lines assumption is met. What do you mean by specific tests David?
The number looks like this Alex 328751430911891.400 or 3.288E+14
What other information do you need?
Thanks for your replies!
Given the small number of regions 0 I would put in a set of dummy fixed effects - one for each region. and see what happens
Also given that there are 7 levels - I presume you mean that the 7 ordered categories on the dependent variable - I would start by treating not as an ordinal variable but as continuous. I also presume that SPSS requires you to sort on the structure of PSUs and that you have done so.
Finally you may want to flip the ordinal scale around so choosing the highest value as the base - this often works if there are not many responses in the lowest category and the ordinal model is using that one with a high standard error as the base.
I did toy with treating it as a continuous DV and running a multiple linear regression but it runs into trouble with the linearity assumptions. I tried to change the order of the ordinal DV but the DEFF is still very high. Do you mean to include regions as a dummy IV?
SPSS merely requires specification of the strata and clusters, then it continues and you can give details about stages and the assumptions about replacement
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS