Difference in difference panel data fixed effects

[Lauren Pin]

URGENT !!
Hi there, I am completely new to stata and have very limited knowledge on econometrics. I have merged two waves of data from the BHPS longitudinal survey (1998) and (1999) and this is my regression model

basic (OLS) DiD model is:

𝑌𝑖𝑡 = 𝛽0 + 𝛽1𝑇𝑖𝑡 + 𝛽2𝐴𝑖𝑡 + 𝜷𝟑𝑻𝒊𝒕𝑨𝒊𝒕 + 𝛽4𝑋𝑖𝑡 + 𝛾𝑖 + 𝜇𝑖𝑡 (2)

where
Yit is the anxiety score in period t for individual i
Tit is a dummy equal to 1 if an individual belongs to the treatment group
Ait is a dummy equal to 1 in the period in which the NMW was introduced
Xit is a vector of individual and job characteristics determining Yit
yi is an unobserved individual specific effect
𝜇𝑖𝑡 is a random error term clustered at the individual level

i have created a dummy variable for the 'treatment' group and a dummy variable called 'year' for the period in which the NMW was introduced
i have generated the interaction term '𝑻𝒊𝒕𝑨𝒊𝒕'

i ran the panel data fixed effects regression
'xtreg anxiety treatment year treatment, fe'

my results are completely insignificant. what am i doing wrong here? is anxiety supposed to be a dummy variable? am i supposed to use a logistic regression?
i am so so confused.

[Phil Bromiley]

You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data. Also, don't say "urgent" - most of the folks who answer questions answer them very quickly but some resent being pushed. We are not generally from your area so many of us have no idea what the BHPS data look like - you need to explain things if they are important.

Putting down exactly what you run is essential. Your xtreg should not work with treatment entered twice and you didn't enter your TitAit variable at all. If you have two years, and the treatment occurred in the second year for all subjects, you can't have both year dummies and treatment dummies.

With only two years and fixed effects, you're estimating more than one free parameter for every two observations. This may be creating your problems. If you can get more years of data, that would almost certainly be a good idea. If the dv is continuous and the scale meaningful, logit is not appropriate. You seldom if ever benefit from taking a continuous variable and making it discontinuous. If the dv is ordinal, you might look at ordinal logit. For interactions, you will find it easier to use factor variable notation (see pdf documentation) - it makes using the margins and marginsplot statements after the estimation much easier.