Hii everyone...
Need your advice regarding type of analysis (in SPSS) for the following:
I want to measure the significant difference in the pattern of a variable (say Y) across different stages of another variable (say X consists of four phases i.e. Xa, Xb, Xc, and Xd). Both the variables Y and X are measured in Likert scale ranging from 1(highly disagree) to 5 (highly agree).
I can't run independent t-test in this account as Xa,Xb,Xc and Xd are not independent to each other.
Kindly suggest for the same. Which statistical approach should I use to establish that variable Y varies across phases Xa, Xb,Xc and Xd significantly (the results may be insignificant also) ?
Regards,
Sayoni
This sounds like a problem for regression; you want to regress Y on X to see if X has a significant effect on Y. Since Y is ordinal you would want to use ordinal regression (or the mixed model version of ordinal regression, if you have repeated measures for X). Since X is ordinal you probably should also have it coded as an ordinal variable, although in my experience ordinal IVs in regression can get complicated, so sometimes it's simpler to just treat it as continuous instead (this is safest when you have a lot of levels; if it's just a 5-point scale that might not be enough).
This sounds like a problem for regression; you want to regress Y on X to see if X has a significant effect on Y. Since Y is ordinal you would want to use ordinal regression (or the mixed model version of ordinal regression, if you have repeated measures for X). Since X is ordinal you probably should also have it coded as an ordinal variable, although in my experience ordinal IVs in regression can get complicated, so sometimes it's simpler to just treat it as continuous instead (this is safest when you have a lot of levels; if it's just a 5-point scale that might not be enough).
The t-test is not suitable for Likert scales, although normally takes place in the literature. In addition to the regression, that Stephen suggests, it is easiest to start with a contingency table to see if there is relationship between different levels from each variable.
Dear Sayoni,
You can use AMOS from SPSS package. Amos program will allow you
Use an Autorregressive model (panel data for longitudinal analysis)
Built the following model in AMOS: Xa--» Xb--» Xc--»Xd--»Y
(Don’t forget to insert error terms in X, Xc, Xd, and Y)
Good luck with your analysis!
I think a 5 x 5 contingency table is the appropriate option if you have two categorical variables and there are five possible values for each variable.
A really great approach to this kind of problem is a permutation test. Permutation tests are non-parametric, and, depending on the implementation, can handle ordinal data as either dependent or independent variables.
For example the coin package in R is quite flexible. I think it is limited to a one-way analysis plus blocks, but I think that would suffice in this case. The documentation is decent, although I don't understand how to select the correct option to handle certain types of data or situations.
Dear all,
I agree with Stephen Politzer-Ahles : you have 2 ordinal variables then you should treat them as ordinals apllying ordinal regression. When both variables are ordinals and the data shows a positive or negative rtendência may apply the linear-by-linear association model. In accordance with SPSS creates the variable interaction given by the variable blade product X and Y column, which is introduced after the covariates janlea loglinear the general model. In the window "factors" include separately the two variables x and y and finally in the custom you include X and Y and the interactive effect..it will be easier to see an apllication...i have it in portugues..sorry..
have a nice time
Helena
- Badges
- Science topic
- Similar topics
- Social Science