I have a data set on autombiles that has a time dimension, cross-sectional dimension with different levels, and multiple observations per cross-sectional unit. The dataset is quasi a panel that is unbalanced, i.e. I have a various number of time observations for different cross-sectional units.

A structure of my dataset is similar to that below (a simplified version):

Time Brand Model   Product_i  Y_it  X_it  Z_t 

t1   Audi  A3      version1   y_11  x_11  z_1 
t1   Audi  A3      version2   y_21  x_21  z_1
t1   Audi  A3      version3   y_31  x_31  z_1

t1   Audi  A4      version1   y_41  x_41  z_1
t1   Audi  A4      version2   y_51  x_51  z_1

t1   BMW   1       version1   y_61  x_61  z_1
t1   BMW   1       version2   y_71  x_71  z_1
t1   BMW   1       version3   y_81  x_81  z_1

t1   VW    Golf    version1   ...         z_1
t1   VW    Golf    version2   ...         z_1
t2   .................................... z_2  
  • Brands and Models repeat over time
  • Products do not repeat over time
  • Time dimension is at monthly level over 13 years (but unbalanced over models)
  • Y_it varies at the product and time level.
  • X_it: some varies over product and time; but! some varies only over brands and time
  • Z_t varies only over time

I want to estimate a simple linear model where Y_it and Z_it are continuous variables, and X_it can contain both contin. and dichotom. variables for each product in each time period.

$$ Y_{it} = const + \beta*X_{it} + \gamma*Z_t + \theta*(X_{it}\times Z_t) + error_{it} $$

My main interest is in $\theta$.

I want to control for time-invariant unobservable variables for cross-sectional units and for time-related unobservable variables (time dummies) with fixed effects in a form of dummy variables.

My general question is "How can I decide what fixed effects and at what level I may include and what variation is then left in explaining the $X_{it}\times Z_t$ relationship?"

More detailed:

  • Am I allowed to include Cross-sectional Fixed effects and Time fixed effects simultaneously in case I have a variable that varies only over time?
  • How can I decide at what level I should have the cross-sectional fixed effects - for Brand or Brand-Model?
  • May I/Should I consider Brand*Time dummies?

Any suggestions / clarifications / links to a relevant reading are appreciated.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.