I think your question is actually a statistics question. You know the impact of an article, but you want to know "what would the impact have been if it had been published differently?" (counterfactual).
A common tool to answer this question is randomization: For example, let's say you want to compare open access versus closed access. What you would do, is you'd randomly publish articles in one or the other access format. This concept is usually used in clinical trials for example, where patients are randomly assigned to a treatment. In the end they compute a simple statistical model with treatment as a covariate to compute the treatment effect.
Of course you can't do that. You can't do a randomized study for paper publishing. In this case you need Causal Inference methods. Causal inference is quite a big field. A good entry is the coursera course: https://www.coursera.org/learn/crash-course-in-causality
You were saying that you already have the data, right? I'd be happy to help if you are interested.