anova and control variables

[GertJan Haenen]

dear all,

I am researching if people have the same willingness to pay to reduce carbon and nitrogen emission. I have created a public goods game, where the money of public fund goes to reduce either carbon, nitrogen or a third hypothetical option (control group). thus I have three groups, one about carbon, one about nitrogen, and i have one control group in which participants receive neutral information.
I have also asked the participants several questions about carbon emissions and nitrogen, such as 'do you think your actions have an effect on total carbon emissions' or 'do you think you are affected by excessive nitrogen emissions'. but halve of these questions only relate to carbon emissions and the other halve only relate to nitrogen emission. so no question/control variable relates to all/multiple group, but only relate to one group.
therefore, my question; if i want to add these questions as control variables do can i add them directly in my ANOVA model or not.
and if not, then how should i use them?

Thanks in advance!

[Carlo Lazzaro]

GertJan:
welcome to this forum.
Please read and act on the FAQ on how to post (more) effectively. Thanks.
That said, I am not clear with what your dependent variable is.

[GertJan Haenen]

GertJan:
welcome to this forum.
Please read and act on the FAQ on how to post (more) effectively. Thanks.
That said, I am not clear with what your dependent variable is.

thanks for the reply. I am indeed new to this forum.
my dependent variable is how many tokens participant keep for themselves (therefore the tokens to the public fund = 10 - tokens kept for themselves). and the independent variable is the information they receive (about carbon/nitrogen emission or neutral) and the purpose of the public fund money (to prevent carbon/nitrogen emissions or a neutral cause). for the following part I will refer to these three group which are my independent variable as carbon, nitrogen or neutral.

my research question is simplified to: is there a difference between the tokens kept for the participants themselves in the three groups (carbon, nitrogen, neutral)

I currently run this Anova in stata: anova resulttotalmedian typetotal occupation gender c.age
- with age converted to a continuous variable
- typetotal is the independent variable about which group they are in: carbon, nitrogen or neutral
- resultotalmedian: ignore the median part, this is my dependent variable how many tokens participants keep for themselves

i have added the three control variables (occupation, age, gender) because they have an effect on all three groups (carbon, nitrogen, neutral)

certain questions in my experiment only refer to one of the three groups (of my independent variable). so certain questions are about nitrogen, and the other questions are only about carbon.
for instance the question: do you think you can have an effect to reduce nitrogen emissions?


my question: I run a anova on three different groups, I have certain variable that only correlate/has an effect on one of the three group, should I still use these variables as a control variable or should I keep them out of my anova.

[Carlo Lazzaro]

GertJan:
1) if your dependent variable takes on integer values, why not considering -poisson-?
I'd figure out something like (considering a square term for -age-):

poisson <depvar> i.gender i.employment c.age##c.age i.group

2) without other details form your side (please follow the FAQ when posting. Thanks), I'd get rid of thode variables which affect one group only.

[Jeff Wooldridge]

GertJan. I wouldn't use ANOVA. You should just use regression analysis. And because your dependent variable is an integer necessarily between zero and 10, I would use, in addition to linear regression, binomial regression. That can be done using the glm command with the family(bin 10) link(logit) option. You can think of, say, "neutral" being the base group. The linear regression would be

reg tokens  i.carbon i.nitrogen i.occupation i.gender c.age, vce(r)

The coefficients on i.carbon and i.nitrogen will give you the difference between those groups and the neutral group. You could include interactions, too. Or

glm tokens  i.carbon i.nitrogen i.occupation i.gender c.age fam(bin 10) link(logit), vce(r)
margins, dydx(carbon) at(carbon = 1 nitrogen = 0)
margins, dydx(nitrogen) at(carbon = 0 nitrogen = 1)