This applies to the case where there is no corresponding author.
Talking about my field (applied CS), where many experiments depends on experiments which are not always easily repeated, and code are usually not available*:
I think it is more effective to go directly to the supervisor (or supervisors) - [last and second to last authors - since this is the order usually followed in applied CS]. These authors usually takes higher care of their reputation in the field. Because of that, they will force their students (i.e. first author and/or second) to give explanation. (example: This happened when my supervisor received a question about a paper of his with another post-doc. I am sure that my supervisor did not work much on the details of the paper, but his name must be included - that s an engineering tradition I believe. The supervisor made sure to call the post-doc and asked for explanation. The supervisor forced the the post-doc to send an email back to the question sender).
If you send a question directly to a PhD/Msc student, he would ignore you most likely - from my experience. Because many PhD students hide their mistakes intentionally - just to have their thesis accepted later. (example: I heard a PhD student that he made some mistakes in his MSc thesis and he hid them intentionally. He said "this is to make whoever work on my work find trouble matching the paper results and the new experiments results".
But honestly, at the end, we cannot generalize.
(*)This one of the biggest mistakes in applied CS.