Si comprendo tu pregunta, en diseño de evaluacion de impacto aunque no tiene un diseño con control equivalente, es general una practica para mitigar esa limitacion tomando los datos de la linea de base. Tienes que verificar condiciones sobre la equivalencia de las muestras. Si los hay, tendrás que revisar el modelo estadistico pertinente (ie, prosensity score, inverse weighted propensity, etc). Si el post cross-sectional tiene grupos no intervenidos, podes revisar la factibilidad de construir el "contrafáctico". Te adjunto este material por si te sirve 

Handbook on Impact Evaluation: Quantitative Methods and Practices (World Bank Training Series) 1st Edition

https://openknowledge.worldbank.org/bitstream/handle/10986/2693/520990PUB0EPI1101Official0Use0Only1.pdf

Suerte y exitos

I think you don't want to include both Change-from-baseline and Baseline in the same model.  I think this would violate an assumption of OLS regression, that the independent variables don't affect each other.

To me, it makes more sense to include the Baseline and Final scores in the model.  This lets you draw conclusions about the Baseline scores, as well as the Final scores.  You have a variable Time that tells you the effect of changing from baseline time to final time.

Consider the following.  Both Red and Green increased about the same amount from Time 1 to Time 2.  But we can also see that Green started out lower than Red.

http://rcompanion.org/handbook/images/image088.png

But if you are really just interested in Change-from-baseline, you can use that as the dependent variable.

By definition, baseline is a control already. It is therefore superfluous to include it as control.