Following the copied link (https://www.statalist.org/forums/forum/general-stata-discussion/general/1323707-fixed-effect-difference-in-differences-model) I was expecting the classic dummy 'Post' to be dropped at the inclusion of year Fixed Effects in a difference-in-difference regression, with panel data (though not balanced).
This is not the case.
Where might it be the error in my data?
Depending on the software you are using, either the "Post" variable, or one additional time dummy (if you're modeling year/month/day fixed-effects), will be dropped. Should you include a "Post" variable and time fixed-effects in a regression model, then R (which I suspect you're using) will automatically exclude an additional time dummy. To illustrate, here is the canonical approach:
$$ y_{it} = \alpha + \gamma Treat_{i} + \lambda Post_{t} + \delta (Treat_{i} \times Post_{t}) + \varepsilon_{it} $$
where $i$ could represent firms or businesses and $t$ could represent months or years. Now, should you wish to estimate this model:
$$ y_{it} = \alpha + \gamma Treat_{i} + \lambda Post_{t} + \omega_{t} + \delta (Treat_{i} \times Post_{t}) + \varepsilon_{it} $$
with $\omega_{t}$ is standing in for "time" fixed-effects (i.e., dummies for each time period), then one additional time period is dropped to allow for estimation. In this setting, only $t-2$ unique estimates for your dummies will appear in the console; one is dropped as normal and the other excluded dummy is likely to show up as NA in your regression output. It should be noted that the variable "Post" will be estimated.
Should you drop the "Post" variable entirely and interact the treatment variable with each time dummy using as.factor notation, then R will only drop one time period for you. In sum, you lose one more time dummy estimating the equation that you referenced.
Below is another question where collinearity problems were addressed.
