Hi,
I use some official data form our national statistical office survey (N = 1350). I have one independent variable of 6 age classes and one dichotomous dependent variable (looking for health related information online 1-Yes, 0-No).
I use SPSS, Chi2 test to detect group differences.
CROSSTABS
/TABLES=IHIF BY AGECLS
/FORMAT=AVALUE TABLES
/STATISTICS=CHISQ
/CELLS=COUNT EXPECTED ROW COLUMN BPROP
/COUNT ROUND CELL.
I have a weight variable. So, when using weights, I get all classes to be statistically significant (attachment 1).
If I don't use weights none of the groups if significantly different from the other (attachment 1).
So, am I using the wrong test, should I just omit weights or is there some other issue with the data?
Thank you for your help.
Your first table has a total count of more than one million. Your second table has a total count of 1350. If your actual N is 1350, I don't know how using the table with N=1226696 would make sense. I don't understand how you are using the weighting variable.
It also may be helpful to look at effect size statistics. For your second table, I get a Cramer's V of about 0.10, which is rather small. You could look at the odds ratio for each age group (using one age group as the reference level).
I'm sorry but I'm a bit confused why aren't you using logistic regression.? Wouldn't that be much easier? Best wishes David Booth
Salvatore, thank you for your thoughts. The weight for individual respondent look like this:
1565,361458000000
1240,320460500000
790,500699180000
1753,826901700000
1356,434749000000
1377,741225200000
2050,964000000000
2050,964000000000
1424,871026000000
1928,870172500000
1405,665912100000
1417,523741900000
1714,363869100000
1385,296605700000
etc.
This is EUROSTAT's explanation of it (because the data is part of the survey Usage of Information and communication technologies in households and by individuals): Individual Weight All individuals IND_WGHT Nnnnnn.nnnnnn - Grossing up factor of the individual (As many digits as necessary. Use decimal point if needed).
So, I suppose SPSS just uses these numbers in some way.
Kind regards
Mitja
Mitja Decman , I still don't understand what the weighting variable means or how it is being used in the analysis.... That being said, the relative proportions for the first table aren't that different from those in the second. (The counts change by a factor of about 900). The p-value ends up very different because of the difference in sample size. Be cautious not to put too much emphasis on the p-value. For either table, the Cramer's V is about 0.1, which suggests a small effect, but might have practical importance. That's up to you.
Why you run chi square test? Under which assumptions? If I’m not mistaken, chi-square is for nonparametric test.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS