Utilisateur
they are decreasing
genom att ta squared. generate a new variable som är upphöjd i 2, men ha kvar den gamla också
att ha med den som är auxiliary variable, det är nör vi generarar en ny variabel som ändå tar vördet av en annan variabel som squared.
x1~ = x1^2 där då x1~ är auxiliary variable
cannot be interpreted as ceteris paribus effects.
the marginal effect would be:
B1 + B2•x1 since we have the model
B1•x1 + B2•x1^2
we can interpret coefficients in models where U≠0 , BUT we cannot compare treatment effects
1% change in x1
=> Y change by B1/100 units
B1 is expressed in percentage, so we have to divide by 100 to get units.
here we can also define a auxiliary variable so we can run regression
x1~ = log(x1)
1 unit increase of x1
=> B1•100 % change in Y
to get expressed as % we have to multiply by 100
we define logged variables in stata by lscore or loutput by putting small L before the variable to denote that we have a logged auxiliary variable.
1 % in x1 => 1% change in Y
benchmark:
when dummy variables = 0
treatment:
when dummy variables = 1
den skiftar interceptet. allt annat lika så kan detta tolkas som treatment effect
man skapar en auliliary variable där man combine at least 2 variables. interaction term is a regressor that is defined as the product of two or more variables
Y= B0 + B1x1 + B2x2 + B3(x1•x2)
marginal effect on x1:
B1 + B3•x2
marginal effect on x2:
B2 + B3•x2