What is ANCOVA?
In a broad term, analysis of covariance (ANCOVA) is equivalent to multiple linear regression. However, there is a difference in the context in which these two approaches are utilized.
In ANCOVA, there is at least one independent variable of interest — the one that is manipulated by the experimenter — and at least one other variable, called a “covariate“. This covariates, i.e., the variable that we are not directly interested in, is treated more like confounding variable. By including covariates we want to make sure that the main source of variance observed for the dependent variable is our experimental manipulation, and not some other accompanying effects. In other words, both kinds of variables: the one that we manipulate with, and covariates, are contributing to the observed variance of the dependent variable. The difference is what was our research question about in the first place.
On the other hand, in multiple linear regression all independent variables are conceptually equal, in the sense that they all contribute to (explain the variability of) the dependent variable. In multiple linear regression there is no distinction between which variable is the experimenter’s manipulation and which is just a covariate (a confound).
The above explanation can be better understood by reviewing this Python example and reading this ResearchGate discussion. The latter I highly recommend, as there is also a thread on why in multiple linear regression interaction between independent variables is not modelled by default (in fact, this is how I stumbled upon this post — I was searching best practices for modelling interaction effects).
Read also: my other post on general linear model and multiple linear regression