Predict conditional mean

[Fredrik Paues]

I would like to predict the mean and standard error conditional on some dummies in a data set containing four variables: y, year, control1 and control2. That is, I would like to estimate E[y | year] for the following linear model y = \alpha + \beta * year + \gamma * control1 + \delta * control2 + \varepsilon. I run the following Stata code:

xi: reg y i.year control1 control2
collapse y control1 control2, by(year)
predict cond_mean, xb
predict cond_mean, stdp

Is this correct? Or am I really doing something entirely/subtly different?

[Carole J. Wilson]

You don't say what version of Stata you are using, but the xi: syntax has been superseded by factor variable notation in more recent Stata versions (see help fvvarlist). Following the regression command, you should use the margin command to get your estimates.

[Weiwen Ng]

I would like to predict the mean and standard error conditional on some dummies in a data set containing four variables: y, year, control1 and control2. That is, I would like to estimate E[y | year] for the following linear model y = \alpha + \beta * year + \gamma * control1 + \delta * control2 + \varepsilon. I run the following Stata code:

xi: reg y i.year control1 control2
collapse y control1 control2, by(year)
predict cond_mean, xb
predict cond_mean, stdp

Is this correct? Or am I really doing something entirely/subtly different?

Note Carole's caution about the -xi- prefix.

Then, -predict- will predict the actual value of -xb- for each person in the data, and no more. I think what you want is probably margins, e.g.

reg y i.year control1 control2
margins, over(year)
margins, at(year = (0 1))

The first -margins- command will present average predicted mean Ys over each value of year. So, for each value of year, -margins- will calculate predicted y, keeping all covariates as is.

The second -margins- command is subtly different. It will show predicted y over all observations, changing their values of year to 0 and 1 respectively. So, it's as if you changed all observations to have year 0 and year 1. You should read up on the Stata manual for margins, or Richard Williams' presentation on margins linked below. (Richard Williams is a frequent poster and he frequently posts here.)

https://www3.nd.edu/~rwilliam/stats/Margins01.pdf