multiple moderated regression steps

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

multiple moderated regression steps

stamkiral
hi,

ım studying moderated effects of percieved social support and justice world belief on relationship between stress coping strategies and depression level. ı haver never run this analysis before soi ı want to check my steps whether correct or not.

first ı run regression
in step 1
centered independent variables and centered moderators

in step2
two way interactions

instep 3
three way interactions

as results ı found significiant two way and three way interactions. "It  ıs important after this"

for example two way interactions;
ı run 2 slopes that fit in with Winnifred's suggestion (http://www.docstoc.com/docs/21151269/Moderated-Multiple-Regression-v5 p.6)
my criteria for significence was the centered indipendent vairable (that has significant interaction level) in Block 2--------- is this way is correct?????

second to plot this slope ı used Dawson' s 2 way unstandardised excel spreadsheet (http://www.jeremydawson.co.uk/slopes.htm). In spreadsheet it is required to enter unstandardised regression coefficients of IV, moderator and interaction. ın which block the coefficients are true for entering IV and moderator co efficients?????
ın 1 block (that include main effects) or in 2 block (that include interactions)?????

one more question: In three way spreadshhet it is necessary to enter variance of IV*Moderator coefficient. ın matrıx which covariance block is correct to enter this value? should ı look at second covariance block or third covariance block.
like this,for value of Covariance of IV*Mod1, IV*Mod2 coefficients which covariance block must be taken as correct???
thaks so much

Note: excuse me for my English
Reply | Threaded
Open this post in threaded view
|

Re: multiple moderated regression steps

Frank Harrell
How does it help to center the variables?  What statistical principles are you using to fit your model and how are you ensuring statistical validity of inferential estimates?  Why do it in multiple steps?  Why use a mixture of software tools?  What is the utility of standardized regression coefficients?
Frank
stamkiral wrote
hi,

ım studying moderated effects of percieved social support and justice world belief on relationship between stress coping strategies and depression level. ı haver never run this analysis before soi ı want to check my steps whether correct or not.

first ı run regression
in step 1
centered independent variables and centered moderators

in step2
two way interactions

instep 3
three way interactions

as results ı found significiant two way and three way interactions. "It  ıs important after this"

for example two way interactions;
ı run 2 slopes that fit in with Winnifred's suggestion (http://www.docstoc.com/docs/21151269/Moderated-Multiple-Regression-v5 p.6)
my criteria for significence was the centered indipendent vairable (that has significant interaction level) in Block 2--------- is this way is correct?????

second to plot this slope ı used Dawson' s 2 way unstandardised excel spreadsheet (http://www.jeremydawson.co.uk/slopes.htm). In spreadsheet it is required to enter unstandardised regression coefficients of IV, moderator and interaction. ın which block the coefficients are true for entering IV and moderator co efficients?????
ın 1 block (that include main effects) or in 2 block (that include interactions)?????

one more question: In three way spreadshhet it is necessary to enter variance of IV*Moderator coefficient. ın matrıx which covariance block is correct to enter this value? should ı look at second covariance block or third covariance block.
like this,for value of Covariance of IV*Mod1, IV*Mod2 coefficients which covariance block must be taken as correct???
thaks so much

Note: excuse me for my English
Frank Harrell
Department of Biostatistics, Vanderbilt University
Reply | Threaded
Open this post in threaded view
|

Re: multiple moderated regression steps

stamkiral
variables were centered (except DV), because in social sciences zero is rarely a meaningful point on a scale (Cohen, Cohen, West, Aiken, 2006). for example in percieved social support questionnaire there is no value as zero. It is a Likert Type questionnaire and on questionnaire 1= strongly yes.....7= strongly no. So Aiken and West (1991), suggested to centered predictors and moderators to appoint zero a meaningful value to count in regression equation. also multilevel steps were run to avoid multicollinearity. according to results .05 were accepted as significant (in the coefficient table) and slope test were done for interactions.
for plottin slopes unstandardised regression coefficient were used.  
Reply | Threaded
Open this post in threaded view
|

Re: multiple moderated regression steps

Frank Harrell
That concern has nothing to do with centering variables before including them in a model.

Your multiple significance testing strategy is not based on statistical principles and will distort all inferences you obtain from the "final" model.
Frank
stamkiral wrote
variables were centered (except DV), because in social sciences zero is rarely a meaningful point on a scale (Cohen, Cohen, West, Aiken, 2006). for example in percieved social support questionnaire there is no value as zero. It is a Likert Type questionnaire and on questionnaire 1= strongly yes.....7= strongly no. So Aiken and West (1991), suggested to centered predictors and moderators to appoint zero a meaningful value to count in regression equation. also multilevel steps were run to avoid multicollinearity. according to results .05 were accepted as significant (in the coefficient table) and slope test were done for interactions.
for plottin slopes unstandardised regression coefficient were used.
Frank Harrell
Department of Biostatistics, Vanderbilt University