I created an experiment with the builder using Pavlovia version 2026.1.1 and ran it on Prolific using the return URL (nothing was coded manually to redirect participants to Prolific). Oddly, a substantial amount of participants produced zero-byte csv files, among those were such who finished the experiment with a code and such aborting the study. I can tell that they actually ran the experiment because 1. some provided the correct code and 2. the start time saved in the name of the respective zero-byte csvs matched the starting time of participants documented by Prolific and Prolific provides the duration of each run for the respective participants. No error codes were reported by these participants, some even described what has happened during the experiment further corroborating that they did actually run it (I even got offers to re-run the experiment). For the majority of participants data was saved normally.
Reading through posts with related titles and tags, I have not found this problem being reported yet. I would be happy to read about ideas on what was potentially going on here.
Sorry to hear you’ve been having issues. Please could you try downloading the data from the project page to see if it matches the data in Gitlab. Do not switch the project to inactive.
Do you have “save incomplete results” and “save periodically” both turned on? (I typically don’t use either myself).
How long is your study (for a typical participant)?
Please could you PM me the URL of the study so we can investigate further (or post if here if you are happy for it to be public)?
This is probably a long-shot, but is there any chance that you have multiple files for these participants and their data is saved in a different file? I ask because I got a bunch of empty files when I ran my experiment, but they all seemed to be extra (so I didn’t look into them further) – I also had complete files for all of my participants that finished the experiment.
thanks for the reply! The data is identical on both servers. I only checked “save incomplete results”. There is quite some variability in how long it takes, since part of it is a pattern learning task, only ending upon reaching a specific criterion - most participants need 35 mins to 55 mins. Here is the URL of the study (there are no credits consumed since I am on a license): https://run.pavlovia.org/severinm18/match_to_sample_causality
Could this problem be attributed to using a specific browser to run the experiment?
Thats possible. Can you think of a mechanism that explains uploading empty files to the server in that case? If this is the explanation, I guess this issue should have already been reported many times..
I ran the study on the 28th of April between 01:15 - 09:00 pm CET. Do you know if there were any server updates or similar things happening at that time?
According to https://status.pavlovia.org/ Pavlovia Gitlab Responsivity (10s) was down for 2 minutes in total on 28th April. However, thank you for the time window.