Occupation, education and college cohort fixed effects

[Samiya Mohammed]

I hope someone can help me with this issue in estimating the below model

The model includes :

• The dependent variable is the proportion of female directors by detailed occupation fields ( 144 field) and by college cohort ( 8 cohorts) .
• The key independent variable is the proportion of females by detailed education fields ( 133 field) and by college cohort (8 cohorts) .
• The control set includes a set of variables .
• I used ordinary least square to examine the association between the key independent variable and dependent variable.
• Standard Errors are clustered by detailed occupation fields ( 144 field) , detailed education fields (133 field) and college cohort to control for within variation in these groups
• I use repeated cross sectional surveys to estimate the above model

My question: do I need to add detailed occupation fields (144 fields) , educational fields (133 field) , college cohort dummies as fixed effects in the model?

[Rhys Williams]

If I understand correctly, you wish to estimate a linear model by OLS of the proportion of female directors against proportion of females and a set of control variables. When you say the variable is "by detailed occupation fields" what does this mean. That you have proportionofdirectors_occupation1, proportionofdirectors_occupation2 ... etc?

Whilst I don't fully understand the dataset you have (perhaps you could provide an excerpt), my instinctual response would be that you want to control for occupation, education and (perhaps) college cohort.

Best,
Rhys

[Samiya Mohammed]

Thank you for your reply Ryhs

You are right I want to control for occupation, education, college cohort , but am confused why i need that ?
I already use the finer categorization to construct the dependent variable and key independent variable of interest

my guess is that i need only to cluster errors by these groups to control the variation within these groups

• The dependent variable : the count of female director within occupation field j divided by the total number of directors in the same occupation j ( same as you explained )

egen total = total(female_director), by(detailed_group_occupation)
bys detailed_group_occupation : gen prop0= (total)/(_N)
drop total

• The key independent variable: the count of female within education field k divided by the total number of individuals in the same educational field k

egen total = total(female), by(detailed_group_education)
bys detalied_group_education : gen prop1 = (total)/(_N)
drop total


Am interested in the association between these two variables

My confusion is whether i need to add occupation and education detailed dummies as fixed effects in the OLS model?

[Rhys Williams]

Thank you for your reply Ryhs

You are right I want to control for occupation, education, college cohort , but am confused why i need that ?
I already use the finer categorization to construct the dependent variable and key independent variable of interest

my guess is that i need only to cluster errors by these groups to control the variation within these groups

• The dependent variable : the count of female director within occupation field j divided by the total number of directors in the same occupation j ( same as you explained )

egen total = total(female_director), by(detailed_group_occupation)
bys detailed_group_occupation : gen prop0= (total)/(_N)
drop total

• The key independent variable: the count of female within education field k divided by the total number of individuals in the same educational field k

egen total = total(female), by(detailed_group_education)
bys detalied_group_education : gen prop1 = (total)/(_N)
drop total


Am interested in the association between these two variables

My confusion is whether i need to add occupation and education detailed dummies as fixed effects in the OLS model?

Thanks for providing more details on the construction of your variables. Yes, I think you ought to control for occupation and education dummies in the OLS model (i.e. add i.occupation and i.education as controls).
At the moment you are effectively running the regression %DirectorsthatareFemale = a + b(%ofstudentsthatareFemale) + e
You are treating each observation as independent (and clustering your standard errors, which is good). But, if there is correlation between %Directors and Education, and %Students and Education (the same for Occupation) then you will have omitted variable bias. Whether correlation in this form exists is something you can test and I encourage you to do so.

best,
Rhys

[Samiya Mohammed]

Thank you Rhys for your very helpful reply and suggestion

[Barc]

Thank you for your reply Ryhs

You are right I want to control for occupation, education, college cohort , but am confused why i need that ?
I already use the finer categorization to construct the dependent variable and key independent variable of interest

my guess is that i need only to cluster errors by these groups to control the variation within these groups

• The dependent variable : the count of female director within occupation field j divided by the total number of directors in the same occupation j ( same as you explained )

egen total = total(female_director), by(detailed_group_occupation)
bys detailed_group_occupation : gen prop0= (total)/(_N)
drop total

• The key independent variable: the count of female within education field k divided by the total number of individuals in the same educational field k

egen total = total(female), by(detailed_group_education)
bys detalied_group_education : gen prop1 = (total)/(_N)
drop total

In addition to these calculations and research, I was also tasked with compiling an annotated bibliography of the literature and sources I use for my research. I had problems with this and I had to go to https://edubirdie.com/annotated-bibl...riting-service where I was given professional help in writing an annotated bibliography.
Am interested in the association between these two variables

My confusion is whether i need to add occupation and education detailed dummies as fixed effects in the OLS model?

I am very grateful to you for the explanation.