Instruction videos
OLS regression
Definitions and SE tests
- Autocorrelation means that a variable (or error term) at time t is influenced by the values of
the same variable (error term) at time t-n
- Heteroscedasticity means that the variance of the error terms is not constant along the
regression line (over time)
- Above both bias the standard errors (and thus significance level), not the beta coefficient!
- Use White standard errors if heteroscedasticity is apparent
- Use Newey-west standard errors if autocorrelation or both are apparent!
Stata commands see video
Panel regression in Stata
- Heterogeneity leads to OVB beta coefficients are incorrect
1
,Probit model
- Use probit if y is dummy!
- Values can only be 0 or 1 (because dummy) so you can not use OLS (because than bigger or
smaller than 1 or 0 is possible)
- Probit: looks at probability if y=1
-
2
, -
Interaction effects
- Test if two independent variables (X) increase or decrease each other’s influence on
dependent vriable (Y)
- Example: [ gender, hours of prep] grade mathematics exam
- Interaction between gender and preparation: male/female students learn more within one
hour of prep.
-
3
, Interaction effect!
B3 is difference in slope coefficient B2 between male and female. B3 negative (because
female slope is less steep)
4
OLS regression
Definitions and SE tests
- Autocorrelation means that a variable (or error term) at time t is influenced by the values of
the same variable (error term) at time t-n
- Heteroscedasticity means that the variance of the error terms is not constant along the
regression line (over time)
- Above both bias the standard errors (and thus significance level), not the beta coefficient!
- Use White standard errors if heteroscedasticity is apparent
- Use Newey-west standard errors if autocorrelation or both are apparent!
Stata commands see video
Panel regression in Stata
- Heterogeneity leads to OVB beta coefficients are incorrect
1
,Probit model
- Use probit if y is dummy!
- Values can only be 0 or 1 (because dummy) so you can not use OLS (because than bigger or
smaller than 1 or 0 is possible)
- Probit: looks at probability if y=1
-
2
, -
Interaction effects
- Test if two independent variables (X) increase or decrease each other’s influence on
dependent vriable (Y)
- Example: [ gender, hours of prep] grade mathematics exam
- Interaction between gender and preparation: male/female students learn more within one
hour of prep.
-
3
, Interaction effect!
B3 is difference in slope coefficient B2 between male and female. B3 negative (because
female slope is less steep)
4
