We randomize wages for census image classification task to estimate discontinuous labor supply elasticities at round numbers---in particular at 10 cents, to test for left-digit bias. We also aim to replicate the upward sloping labor supply functions to a given task estimated in Horton et al (2011).
External Link(s)
Citation
Dube, Arindrajit, Alan Manning and Suresh Naidu. 2017. "Left-digit Bias in Wage Perceptions and Labor Supply in an Online Labor Market." AEA RCT Registry. March 01. https://doi.org/10.1257/rct.1349-2.0.
Indicator for accepting MTurk task. Time to complete task. Error rate in tagging.
Primary Outcomes (explanation)
Error rate will be calculated using 2 "known positives" that will be randomly assigned to workers.
Secondary Outcomes (end points)
Secondary Outcomes (explanation)
Experimental Design
We will post a total of 5,500 unique HITS on MTurk tasks for $0.10 that includes a brief survey and a screening task, where respondents view a digital image of a historical slave census schedule from 1850 or 1860, and answer whether they see markings in the “fugitives” column. Some respondents are offered a choice of completing an additional set of classification tasks for a specific wage.
Extension: we have also run a version of this experiment with a recipe tagging task, at 90, 99, 100, 101, and 110 cents.
Experimental Design Details
We refer to the initial screening part as “stage-1.” Those who complete stage-1 and indicate that the primary reason for participation is "money" or "skills" (as opposed to "fun") are then offered an additional task of completing either 6 or 12 such image classifications (chosen randomly) for a specific (randomized) wage, w, which we refer to as "stage-2." If they accept the stage-2 offer, they are provided either 6 images (task type A) or 12 images (task type B) to classify, and are paid the wage w. These 5,500 HITs will remain posted until completed, or for 3 months, which ever is shorter. Any single individual on mTurk (identified by their MTurk ID) will be allowed to only do one of the HITs. We aim to assess the left-digit bias in wage perceptions experimentally by randomizing the offered wages for HITs on MTurk by randomizing a wage offer for a HIT to vary between $0.05 and $0.15, and assessing whether there is a jump in the acceptance probability between $0.09 and $0.10 as would be predicted by a left-digit bias.
Randomization Method
Randomization done by computer that automatically posts task on Amazon turk.
Randomization Unit
individual MTurk worker.
Was the treatment clustered?
No
Sample size: planned number of clusters
5500
400 for extension
Sample size: planned number of observations
5500
400 for extension
Sample size (or number of clusters) by treatment arms
5500
400 for extension
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)