help
help > RE: Testing an interaction: split or not & issues with "0" coding
Jun 18, 2020 07:06 PM | Alfonso Nieto-Castanon - Boston University
RE: Testing an interaction: split or not & issues with "0" coding
Dear Lucas,
You may simply add covariates of no interest to your design (e.g. "males, females, score*males, score*females, age, site1, site2") and enter 0 at the corresponding place in your contrast (e.g. [0 0 -1 1 0 0 0]). If, in addition, you want to also correct for potential covariate-by-group interactions, that is also perfectly fine, and you may do so by entering those interaction terms in the design as additional covariates of no interest (e.g. "males, females, score*males, score*females, age*males, age*females, site1*males, site1*females, site2*males, site2*females" and a contrast [0 0 -1 1 0 0 0 0 0 0]). Of course, you may also do that differentially across covariates (e.g. include the group*age interactions but not the group*site interactions)
There is no potential confusion regarding continuous vs. categorical covariates here. In both cases you are estimating the expected differential effect in connectivity associated with a unit-increase in scores when everything else (including the covariates) remain constant (i.e. at any arbitrary value of those covariates, whether continuous or categorical), and you are then simply comparing those differential effects between your two groups (males vs. females).
Hope this helps
Alfonso
Originally posted by Lucas Moro:
You may simply add covariates of no interest to your design (e.g. "males, females, score*males, score*females, age, site1, site2") and enter 0 at the corresponding place in your contrast (e.g. [0 0 -1 1 0 0 0]). If, in addition, you want to also correct for potential covariate-by-group interactions, that is also perfectly fine, and you may do so by entering those interaction terms in the design as additional covariates of no interest (e.g. "males, females, score*males, score*females, age*males, age*females, site1*males, site1*females, site2*males, site2*females" and a contrast [0 0 -1 1 0 0 0 0 0 0]). Of course, you may also do that differentially across covariates (e.g. include the group*age interactions but not the group*site interactions)
There is no potential confusion regarding continuous vs. categorical covariates here. In both cases you are estimating the expected differential effect in connectivity associated with a unit-increase in scores when everything else (including the covariates) remain constant (i.e. at any arbitrary value of those covariates, whether continuous or categorical), and you are then simply comparing those differential effects between your two groups (males vs. females).
Hope this helps
Alfonso
Originally posted by Lucas Moro:
Dear Alfonso
Dear Conn-Users
Previous posts suggest the following 2nd level contrast [0 0 1 -1] for testing the gender by score interaction, using "males" , "females", "score_for_males", "score_for_females", respectively. However, how do you test the same interaction, by simultaneously correcting for covariates of no interest, such as age or sites? Doesn't it confuse the Conn with categorical or continuous variables when you put "0" for these e.g. [0 0 0 0 0 1 -1]? I also assume here the scores are coded with either a value or a "0". Does it additionally require spliting each covariate into "age_males", "age_females", "site_1_male", "site_1_female", "site_2_male", "site_2_female"?
I screened the posts for this specific question, but could find no answer and would appreciate any comment. Thank you.
Greetings,
Lucas
Dear Conn-Users
Previous posts suggest the following 2nd level contrast [0 0 1 -1] for testing the gender by score interaction, using "males" , "females", "score_for_males", "score_for_females", respectively. However, how do you test the same interaction, by simultaneously correcting for covariates of no interest, such as age or sites? Doesn't it confuse the Conn with categorical or continuous variables when you put "0" for these e.g. [0 0 0 0 0 1 -1]? I also assume here the scores are coded with either a value or a "0". Does it additionally require spliting each covariate into "age_males", "age_females", "site_1_male", "site_1_female", "site_2_male", "site_2_female"?
I screened the posts for this specific question, but could find no answer and would appreciate any comment. Thank you.
Greetings,
Lucas
Threaded View
| Title | Author | Date |
|---|---|---|
| Lucas Moro | Jun 15, 2020 | |
| Alfonso Nieto-Castanon | Jun 18, 2020 | |
| Lucas Moro | Jul 2, 2020 | |
| Alfonso Nieto-Castanon | Jul 6, 2020 | |