I know this is probably very simple but I’m struggling to get my head around this. I want to determine the impact of female teachers (as a count of female teachers) on gender equality ( a score but not a test score) compared to male teachers – the hypothesis is that more female teachers will increase gender equality. I am aware there are problems with endogeneity etc but I just want to focus on the data analysis part for the moment.
(This is not the actual data I'm using, I've changed it.The data I'm using is panel data at the country level so total female teachers in the country and gender equality score for the country over time - again I am aware of endogeneity issues etc)
When I do a single regression with panel data the impact of female teachers on equality is insignificant as is the impact of male teachers and total teachers when I run them as a simple single regressions. However, when I run a multiple regression with female and male teachers both become significant with opposite signs for their coefficient. When I include female teachers, male teachers and total number of teachers only female teachers is significant. I mostly understand why this is but I’m just not sure which result is the most reliable – the single, male and female or male, female and total since they are all quite different.
Any help is greatly appreciated, thank you for your time
This is not simple because the variable is bounded by 0 and 70 and not continuous. Different choices are available for type of regression. See the attached search. Now why do you think this is panel data? BTW you should consider all of this before you collect data. Best wishes David Booth
I don’t fully understand what you are doing but would be surprised that you would need the three absolute counts, or indeed any absolute count.
so I would use either proportion female or proportion male but not both as that would elicit exact collinearity. You would anticipate that a negative coefficient for one would be positive for the other as you are dealing with a closed ratio.
I don't fully understand what you are doing either. The way I imagine it, occasions of observation are at Level 1, and they are clustered within individual students, who are at Level 2. If observations are made every school term, for example, the proportion of (fe)male teachers a particular student has each term would be a Level 1 variable. And characteristics of the students (e.g., sex) are Level 2 variables. Is this characterization of your data reasonably accurate? If not, please clarify. HTH.
I would never have guessed that is what you are doing!
I am still concerned with the use of absolute numbers of females as that will reflect countries with a bigger population and not the preponderance of women in that role. But you know your research questions and I do not. The variable will also (I suspect) be close to unchanging over time, that is made up mainly of between variation. On that issue have a look at Article Explaining Fixed Effects: Random Effects Modeling of Time-Se...
and Article Fixed and Random effects models: making an informed choice
I was suggesting that you use the percentage of workers in that role who are female (that is number of Females in role/ (number of Males plus Females in that role).
You could use percent who are male , but you cannot include percent male and percent female at the same time as they will always be correlated -1 , that is exact collinearity.
- Badges
- Science topic
- Similar topics
- Medicine
- Disease
- Infectious Diseases
- Immunization
- Vaccination