Hello everyone. We have been analyzing the data collected from the previous trial, and we are very pleased with the results! Thank you for all those who accepted our HITs, we learned a lot from the first trial and are excited to release a second one later this week (Wednesday or Thursday).
Here are some details for the upcoming second trial:
- $0.15 per assignment
- No bonuses
- 5 assignments per HIT
- 15 minute time limit
- Same HIT template as previous trial, including example images.
- Same stock qualifications as the last trial:
- HIT Approval Rate greater than or equal to 95%
- Location is one of AU, CA, NZ, GB, US.
- We will be dropping the Number of Hits Accepted Qualification to 5000.
- We will NOT be rejecting any submissions for this trial.
- 49 individual HITs will be released over the 7 day trial.
- HITs will be published at the same times as the previous trial:
- 6:00 am, 10:00 am, 2:00 pm, 4:30 pm, 5:00 pm, 5:30 pm and 9:00 pm.
The wage will be the same as the first trial, except we will not be giving bonuses. This was a means for us to reach out and provide individual feedback, rather than to reward exemplary performance. This method proved too prone to human error, where some workers would be skipped and others would receive more than 1 bonus per submission. For this next trial we encourage all workers to reach out to us through the forums or direct e-mail if they would like some clarification or feedback, or to provide us with feedback.
We will be setting the time limit to 15 minutes to allow for the large images to load completely in the MTurk stock UI. Once time permits we'll move the images to a separate server to cut down on image loading time.
Also, in the previous HIT we specified that there were different types of smoke we would be looking for. In this trial we want to clarify that we want workers to click on images that contain ANY smoke whatsoever.
After reviewing the data we decided not to go forward with a custom Qualification yet. The same stock qualifications as last time will be used, with the exception of dropping the number of HITs accepted to 5000. We understand that some very interested people may have a hard time accessing our HITs because of this. Our priority is on accuracy and speed, and feel that both these priorities would be better met if we avoided custom qualifications at this time.
No submissions will be rejected, so do your best without fear of damaging your reputation with our HITs.
The HITs will be released as new HITs each time to make organization and data entry easier for us. The naming convention for our HITs will look something like: HBW Trial 2 Day 1 Batch 1. The HITs will be released at the same times as the previous to make the two trials comparable for our future analyses.
Here are some questions that we have:
How do we reach out to workers once we have their worker IDs?
We are interested in finding out where people live that accept our HITs, is this something that can be done with worker IDs?
Is 15 minutes long enough to compensate for the image loading times?
Were the smoke examples provided sufficient or would you benefit from a more robust smoke example image dataset?
Would it be beneficial to you to include examples in the forums as well as the HITs themselves?
Are many workers full time mechanical turkers, or is the vast majority of workers accepting HITs in their spare time?
We appreciate your continued support and feedback as we continue to learn and develop our process. Thanks to all those who have already provided feedback and guidance, and we look forward to hearing more from you!
HBW