Do you want to consider the pre-test as a covariate or as a moderator?  I agree with Guillermo Enrique Ramos's suggestion to use the post-test as a response variable rather than a gain score (as an ANCOVA design would do).  But if you want to test for moderation, you may need to include an additional categorical variable based on groups of the pre-test (low, middle high, for instance) to see if there is a significant interaction between your pre-test level and the treatment (a moderating impact).  If no interaction, then you have a simple analysis of covariance.

Dear Ludmila

This is a matter of covariance analysis (ANCOVA) where the concomitant variable (pre-test) should not be affected by the treatment, if the pre-test are independent of the treatment assigned, then the pre-test would be a good covariable.

However I think that it may be better to use a model that explains the post-test, as response variable, in function of pre-test and group (two groups: experimental vs. control), because the interpretation of the results might be more evident.

Meanwhile, if the response variable was the difference, then the contribution of pre-test as a concomitant variable, would be trivial (the pre-test was used to calculate the response), but it would be difficult to distinguish whether this contribution is only due to its application in the formula (with the minus sign), or if it really helps to explain the results.

dear Ludmila, maybe one way to take into account the pre.test value is to measure the difference between maximum score achievable and baseline at the denominator in a fraction whose numerator is the post-pre treatment effect

Hoping of having been useful, Mauro

Do you want to consider the pre-test as a covariate or as a moderator?  I agree with Guillermo Enrique Ramos's suggestion to use the post-test as a response variable rather than a gain score (as an ANCOVA design would do).  But if you want to test for moderation, you may need to include an additional categorical variable based on groups of the pre-test (low, middle high, for instance) to see if there is a significant interaction between your pre-test level and the treatment (a moderating impact).  If no interaction, then you have a simple analysis of covariance.