Effects coding

[Alan James McMichael]

Dear STATA users.

I am currently running a discrete choice experiment with treatment choice as the dependent variable and the headaches, chest tightness, FEV1 and lung exacerbations as the independent variable. Each of the independent variables has three levels, for example headaches has "low" "moderate" and "severe". At present I have the variable headaches with 1=low, 2=moderate and 3=severe. My understanding is that this is dummy coded. However, I have recently read that effects codon is much better for discrete choice. But I am unsure of how to do effects coding in STATA. Do I create new variables?

Any help would be appreciated

Best wishes

Alan

[Marcos Almeida]

Hi Alan,

There is an interesting thread on the subject:http://www.statalist.org/forums/foru...ssion-in-stata

Please prefer to use Stata instead of "STATA".

Hopefully it helps.

Best,

Marcos

[Roman Mostazir]

Hi Alan,

You can use any coding scheme as long as you know what comparison you are doing and how to interpret the regression coefficients. For example, if you have an effect coded independent variable with two categories, your regression coefficient needs to be multiplied by 2 in order to achieve a dummy coding equivalent regression coefficient. Effect coding scheme is particularly used in some special circumstances i.e. in Fractional Factorial designs when you want to estimate the main effects without any collision with the intercept terms. Also, if you have several categories in your independent variables, effect coding helps to estimate the main effects and interaction effects while with dummy coding the main effect will represent 'simple main effect'. Two papers below are good read to understand the mechanism and there are of course others. And a precaution, effect coding takes -1 in the coding scheme which Stata will flag as impermissible to be used as factor variable (help fvvarlist). The overall execution of effect coding and tests for effects may depend on post estimation commands using 'margins' and 'contrasts'.

1. E¡ects coding in discrete choice experiments, Mickael Bech and Dorte Gyrd-Hansen, 2005

2. Effect coding versus dummy coding in analysis of data from factorial experiments, Kari C. Kugler, Jessica B. Trail, John J. Dziak, Linda M. Collins, The Pennsylvania State University Technical Report Series #12-120