I am currently attempting to estimate the relationship between writing improvement (Y) and feedback students received. To do this, I have collected data of feedback and writing scores of first and final drafts of a writing assignment (i.e., pre scores and post scores) for 300 students. Every student in my study received both highlevel and lowlevel feedback on their first draft. So my regression equations are as follows:
equation 1: Y= a+B1X1+B2X2+e (for high-level) (Y: post high-level score, X1: pre high-level score, X2: number of high-level comments)
equation 2: Y= a+C1X3+C2X4+u (for low-level) (Y: post low-level score, X3: pre low-level score, X4: number of low-level comments)
The reviewer said that it’s not enough to conduct analysis separately for two groups (highlevel and lowlevel) but rather equality of regression coefficients should be tested explicitly through multigroup analysis or via interaction terms. I pooled these two datasets (N=600) and tested coefficients equality via interaction term (adding a dummy variable, feedback focus, highlevel or lowlevel. I have centered the number of comments (highlevel and lowlevel, they are in the same column), but still failed to get rid of multicollinearity.
I have read a bit on the Chow test, but I am not quite sure whether it can be used in my study. In the studies using Chow test, the samples are split into two groups by an indicator variable such as gender, male and female. Male and female are exclusive. But in my study, feedback focus, highlevel and lowlevel, feedback focus, are not exclusively divided. Every student in my study received both highlevel and lowlevel feedback on their first draft. In other words, the sample students are exactly the same for both high-level and low-level groups.
I would really appreciate any help in doing this. Thank you very much in advance.
equation 1: Y= a+B1X1+B2X2+e (for high-level) (Y: post high-level score, X1: pre high-level score, X2: number of high-level comments)
equation 2: Y= a+C1X3+C2X4+u (for low-level) (Y: post low-level score, X3: pre low-level score, X4: number of low-level comments)
The reviewer said that it’s not enough to conduct analysis separately for two groups (highlevel and lowlevel) but rather equality of regression coefficients should be tested explicitly through multigroup analysis or via interaction terms. I pooled these two datasets (N=600) and tested coefficients equality via interaction term (adding a dummy variable, feedback focus, highlevel or lowlevel. I have centered the number of comments (highlevel and lowlevel, they are in the same column), but still failed to get rid of multicollinearity.
I have read a bit on the Chow test, but I am not quite sure whether it can be used in my study. In the studies using Chow test, the samples are split into two groups by an indicator variable such as gender, male and female. Male and female are exclusive. But in my study, feedback focus, highlevel and lowlevel, feedback focus, are not exclusively divided. Every student in my study received both highlevel and lowlevel feedback on their first draft. In other words, the sample students are exactly the same for both high-level and low-level groups.
I would really appreciate any help in doing this. Thank you very much in advance.
Comment