I learned that many researchers show categorical effect in moderation analysis like age, experience and suggest to make dummy variables of these categorical variables. Once one makes dummy variable it creates many variables less than one to the total number of category, its mean that there may be many dummy variables but while looking at the result there one can find only one variable named age or experience whose effect is shown on other variables. Would someone clarify it?
Muhammad, I am not sure whether I understood you correctly. If you have a continuous variable, you should not categorize because you are losing information, and Type I error rate may be inflated. On the other hand, if you have a dichotomous variable, than you have to create a dummy variable which is then included in your moderator or mediator model. If you have more than two categories, then the number of dummy variables will be the number of categories minus one. For example, if you are measuring the happiness of people in Asia, Arabia, and Europe, you need two dummy variables and you have to define the reference group, e.g., europe.
The interpretation of the regression coefficients are then:
The intercept is the mean of people belonging to the reference category, i.e., the mnean of people in Europe. The coefficient of the first dummy variable would be the mean difference in happiness between people belonging to Asia vs. Europe. The coefficient of the second dummy variable would be the mean difference in happiness between people belonging to Ariabia vs. Europe.
Here is a nice explanation on how to interprete dummy variables:
http://essedunet.nsd.uib.no/cms/topics/regression/5/2.html
Regards, Karin
Muhammad, I had a look at the article and I think I now understand your question better. In this study, the authors used two dummy variables: gender was coded 1 for male, 2 for female, ethnicity was coded 1 for non-white, 2 for white. Age is continuous, as you can see in Table 1.Therefore all dummy variables had only two categories, resulting in a single dummy variable for each construct.
In the regression analysis (as well as in the correlation matrix), the authors used gender and ethnicity as labels for their dummy variables. This is of course quite misleading. Gender should be re-labled as 'female' (because females get the higher value), and ethnicity should be re-labled as 'white'. Actually it would have been better to code both variables 0,1, as the results of parameter estimates in multiple regression would have been easier to interpret.
Does this help?
Regards, Karin
Muhammad, If it is possible refrain from dummy coding as creating binary variables (with two answers) may lead to losing important information. For example, if it pertains to age with a cut off point of 35 years someone aged 18 and 34 will be in one group (younger) and people aged 34 and 35 will be separated. Hope that helps a bit.
- Badges
- Science topic
- Similar topics
- Correction