Artificial dataset to satisfy graduate course requirement.
Found on Kaggle.
This dataset was generated in order to fullfill a requirement for a graduate class in applied econometrics. I originally wanted to collect data on the effect of COVID-19 on student performance from a school district, but was unable to given that our local district was already conducting their own research.
The set contains a panel dataset, meant to emulate 6 semesters/trimesters with the first three taking place before the COVID-19 lockdowns, and the final three coming after the lockdowns. It also contains a cross-sectional dataset that is meant to be a single semester/trimester after the COVID-19 lockdowns. Variables were included and manipulated to model real world trends, or local demographics in Portland Oregon. There is a full list of variables at the end of this markdown.
It should be noted that student performance has greatly been diminished as a result of online education.
The Stata code was immense and I thought that it might prove to be interesting/helpful to someone else.
main.do
├── generatedata.do
├── generatepaneldata.do
├── realresults.do
├── panelrealresults.do
├── visuals.do
└── regress.do
- generatedata.do - creates variables and generates relative to chosen distribution
- generatepaneldata.do - creates panel dataset by running generate.do n times
- realresults.do - creates trends in the cross-sectional dataset
- panelrealresults - creates trends in the panel dataset
- visuals.do - prints a small set of graphics that may be interesting in visually checking things like multicollinearity or autocorrelation
- regress.do - runs a pre-defined list of regressions
Each program can be run quietly or manipulated in any way desired, but starting in the generatedata.do file may be most beneficial.
All of the variables were generated and manipulated to model real world trends and local demographics in Portland, Oregon during the 2020-2021 timeline. There were several categories of information to consider, and they're divied up between Personal Information, School Performance, and State Performance. The idea was that state level examinations would be less liekly to have grade inflation, and so may be a more accurate measure of performance.
studentID was used mainly as a means to efficiently differentiate students. The first 700 are in school #1 (the impoverished school), and the rest are in school #2 (the wealthy school). In addition, gradelevel and gender easy to imagine uses for. covidpos is meant to show a negative relationship between having the Covid-19 virus and school performance, assuming time spent off recovering. timeperiod is simply the number of semesters/trimesters from the beginning of the pandemic and online learning. For the panel dataset, the first three time periods are before the pandemic beginning, and the last three from after the switch to online learning.
- readingscore
- writingscore
- mathscore Three "tests" were created to measure performance in macro subject areas. A reading, writing, and math exam could then be good indicators and ways to measure effects of other variables.
- readingscoreSL
- writingscoreSL
- mathscoreSL The tests are considered at the state level as well.
Scores use a bimodal distribution with equal weight and relatively close in means. All others have a normal distribution fitted to means and apprpriate variances given population characteristics in Portland, Oregon.
Creating trends was simple, for example, showing that mean perform worse at the school level, but better on the state level.
replace mathscore = (mathscoreSL - 1) if school == 1
replace readingscore = (readingscoreSL - 1) if school == 1
replace writingscore = (writingscoreSL - 1) if school == 1
replace mathscoreSL = (mathscoreSL - 1) if school == 1
replace readingscoreSL = (readingscoreSL - 1) if school == 1
replace writingscoreSL = (writingscoreSL - 1) if school == 1