Getting help with posting a project on Mechanical Turk
March 15, 2017 7:27 PM   Subscribe

Help me find the most efficient way to accurately convert about 500 pages of charts (in which approximately 60 percent of the data is handwritten) into a spreadsheet.

I have sales logs that span about ten years that I'd like to transfer into a spreadsheet. Each entry contains an identification number (i.e, A11 4343455), sales date, individual's name, and sales price. About six years of the data is handwritten, so it will need to be manually input.

My research suggests that Mechanical Turk is likely the best option, but I have not used it before and I have about a week to get this done. I'd like to work with someone who knows how to ensure that the job gets done correctly, but I'm not having much luck finding references, etc. Any suggestions or pointers would be greatly appreciated, along with any alternatives to Mechanical Turk.
posted by ajr to Computers & Internet (5 answers total) 4 users marked this as a favorite
 
This doesn't solve the scale problem, but DataThief is made for this kind of work.
posted by rachelpapers at 7:59 PM on March 15


Metafilter Jobs?
posted by Night_owl at 8:20 PM on March 15


One approach I've seen used is to hire two (or more) people via Mechanical Turk and have them both do all the data entry. You can then compare the results against each other as a quality check.
posted by une_heure_pleine at 9:52 PM on March 15 [1 favorite]


I've used Mechanical Turk before, and it would certainly work for what you are trying to do. Some things to consider:

- Ethical treatment of workers. Many workers rely on MTurk for some or all of their income, but unfortunately some requesters view MTurk as a way to get a lot of work done very cheaply and set rewards that work out to be far below a living wage. I recommend estimating how long each task will take to complete (or better yet, have someone else do a few and time them) and setting the reward at a level that works out to a living wage for the amount of time the task takes. If you can't afford to pay that much, you shouldn't use Mechanical Turk.
- Quality. Workers on MTurk do many tasks as quickly as possible, since they're paid by the task not by the amount of time spent on it. Quality can vary. You can set up your tasks so that only workers with a high rating can do them, and this helps a bit. As tasks are completed you can review them and accept/reject. You could check everyone's work before deciding whether or not to pay them, but this might be nearly as onerous as doing it yourself. And if you're very strict about mistakes, you'll end up with (understandably) angry workers who will give you bad reviews on Turkopticon. Having multiple people do each log and comparing could be a good way to handle the quality issue.
- Ease of use. Setting up tasks on MTurk is not terribly complicated, but it's not a breeze either. Your project is probably one of the easier types of tasks to set up on MTurk though, and downloading the data at the end is quite easy.
posted by thrungva at 10:13 PM on March 15 [3 favorites]


Setting up a custom job for a portion of the work on with a one-day turnaround for part of the work would allow you to vet whomever you hire. Fiverr, Upwork, or Elance would work. The last two both have teams working together so you could theoretically get more done faster.

Of course, I'll volunteer for $50/hour. And I'm sure I'm not alone. :)
posted by Kalatraz at 3:29 AM on March 16 [1 favorite]


« Older How much protein and carbs do I need right after a...   |   This is all very puzzling to me Newer »

You are not logged in, either login or create an account to post comments