I was at a conference a couple of years ago and a senior colleague, one who I deeply respect, summarized the conversation as: “our labor data are crap.” I think he meant that we have a general problem when looking at labor productivity (for agriculture in this case) both in terms of the heroic recall of days and tasks we are asking survey respondents for, but also we aren’t doing a good job of measuring effort.
For the latter problem, an interesting new paper by Oladele Akogun, Andrew Dillon, Jed Friedman, Ashesh Prasann and Pieter Serneels, gives me some hope. This paper is a follow on piece to their earlier work on malaria testing and treatment, which I blogged about some time back.
The setting is (still) a sugarcane plantation in Nigeria. This setting gives them a lot – sugarcane is a really measurable output – workers cut “rods”, stalks of cane, which are a standard size. The output is paid on a piece rate, so both the plantation and the worker keep careful track of the number of rods cut. And, finally, workers are bussed in and out of the plantation, so there is a regular, fixed work day.
Into this setting, comes a health experiment. Akogun and co. are testing these workers for malaria, which is endemic in this area, and treating them when they are found to be positive. They use the random roll out of this testing and treatment to identify effects (there’s a longer description in my earlier post).
Measurement
One of the central contributions of this paper is to help us understand how hard people are working. Now it turns out the gold standard for this in fields outside of economics is the intriguingly named “double labeled water method” (and if you are really interested, I found a description here). Suffice it to say, it’s expensive and it sure as heck won’t be easy to implement in rural sub-Saharan Africa.
So what’s a development economist to do? Akogun and co. use accelerometers, in this case in the form of belt mounted FitBits. These FitBits measure four levels of activities: sedentary and lightly, fairly, and very active. Using “a proprietary algorithm” the FitBit assigns these levels for a given minute. And, in case you were worried, the workers cannot turn them off.
Akogun and co. assign these FitBits to a sub-sample of 83 people. Each device is uniquely numbered, so workers can’t switch them around. Once a week, the research team downloads the weekly data and recharges the FitBit for the next week.
So how well does the FitBit measure effort? Before turning to the results, it’s worth noting that Akogun and co. run into a bunch of issues with implementing this new measurement tool, and I’ll return to some of those below. At this juncture, it’s worth noting that they are pretty careful about dealing with these problems in their interpretation and analysis of the data.
And so the short answer to the question is yes. Recall that Akogun and co. are working with workers where both the days worked and the actual amount harvested (translated into earnings at a constant price per piece) are measured. Any level of FitBit measured activity is associated with the likelihood that workers showed up on the plantation to the tune of one additional hour of activity correlating with one additional day of work. In terms of levels of activity, an additional hour of light activity increases the probability of a day of work by 10 percentage points, moderate by 7.5 and heavy by 6 percentage points. In terms of earnings, moderate activity seems to be the big (but imprecisely estimated) winner. Given that the Fitbits are calibrated to see heavy activity as all-out running, this makes sense. Putting together unconditional and conditional estimates, Akogun and co. argue that that about 60 percent of the activity association with earnings is driven by showing up for work and 40 percent by more intense activity.
What were the measurement issues? It turns out that Nigerian sugarcane workers, just like me, lose their Fitbits. Of the 83 that start with the devices, 25 have lost them by the end of the study (and no observable variables predict the loss). This seems to be fairly common among the (few) studies that have tried to use these – Akogun and co. note that other studies have found up to 40 percent of individual-day data to be not usable. Akogun and co. close with the argument that further work needs to be done to see what can be done to increase compliance with measurement. They also raise a second issue/question: where is the optimal place to clip the FitBit. They note that it may be worth comparing the use of multiple accelerometers versus singles to tease this out.
And Malaria
The malaria results are interesting (as they were in the original paper). Akogun and co. have set up the experiment so that they can measure a) the simple impact of offering malaria treatment and testing (the ITT), b) the impact of being tested and found to be positive and treated (ToT), and c) the impact of being tested but not infected (treatment on the medically untreated). Now, the plantation shut down for a couple of weeks during their experiment, so they have some power issues, but with the sample they do have they find some changes in labor patterns. Overall testing leads to a shift from light to fairly active effort. Conditional on working, the effects are stronger, with a similar shift out of light activity to fairly or very active activity on the order of 0.8 hours per work day. This increases output by about 25 percent.
For those tested and found to be positive for malaria, the shift out of light activity (again conditional on working) is larger – about double the ITT estimates. This makes sense – when you get treated for malaria, you are more productive than folks who are sick and not treated. However, in line with their earlier paper, there is also an effect for folks who aren’t sick and find out the good news. Conditional on working, these folks are also experiencing a significant increase in fairly active hours, and this seems to be coming mostly from a decline in sedentary hours. So finding out you’re healthy seems to motivate folks to work harder.
This paper shows us a neat way to measure actual effort when we try and measure labor. By doing it in a setting where output and hours worked are cleanly measured, Akogun and co. have given us some clear confidence that this measure works. It clearly has some way to go on adherence and maybe precision, but it’s a big step forward.
For the latter problem, an interesting new paper by Oladele Akogun, Andrew Dillon, Jed Friedman, Ashesh Prasann and Pieter Serneels, gives me some hope. This paper is a follow on piece to their earlier work on malaria testing and treatment, which I blogged about some time back.
The setting is (still) a sugarcane plantation in Nigeria. This setting gives them a lot – sugarcane is a really measurable output – workers cut “rods”, stalks of cane, which are a standard size. The output is paid on a piece rate, so both the plantation and the worker keep careful track of the number of rods cut. And, finally, workers are bussed in and out of the plantation, so there is a regular, fixed work day.
Into this setting, comes a health experiment. Akogun and co. are testing these workers for malaria, which is endemic in this area, and treating them when they are found to be positive. They use the random roll out of this testing and treatment to identify effects (there’s a longer description in my earlier post).
Measurement
One of the central contributions of this paper is to help us understand how hard people are working. Now it turns out the gold standard for this in fields outside of economics is the intriguingly named “double labeled water method” (and if you are really interested, I found a description here). Suffice it to say, it’s expensive and it sure as heck won’t be easy to implement in rural sub-Saharan Africa.
So what’s a development economist to do? Akogun and co. use accelerometers, in this case in the form of belt mounted FitBits. These FitBits measure four levels of activities: sedentary and lightly, fairly, and very active. Using “a proprietary algorithm” the FitBit assigns these levels for a given minute. And, in case you were worried, the workers cannot turn them off.
Akogun and co. assign these FitBits to a sub-sample of 83 people. Each device is uniquely numbered, so workers can’t switch them around. Once a week, the research team downloads the weekly data and recharges the FitBit for the next week.
So how well does the FitBit measure effort? Before turning to the results, it’s worth noting that Akogun and co. run into a bunch of issues with implementing this new measurement tool, and I’ll return to some of those below. At this juncture, it’s worth noting that they are pretty careful about dealing with these problems in their interpretation and analysis of the data.
And so the short answer to the question is yes. Recall that Akogun and co. are working with workers where both the days worked and the actual amount harvested (translated into earnings at a constant price per piece) are measured. Any level of FitBit measured activity is associated with the likelihood that workers showed up on the plantation to the tune of one additional hour of activity correlating with one additional day of work. In terms of levels of activity, an additional hour of light activity increases the probability of a day of work by 10 percentage points, moderate by 7.5 and heavy by 6 percentage points. In terms of earnings, moderate activity seems to be the big (but imprecisely estimated) winner. Given that the Fitbits are calibrated to see heavy activity as all-out running, this makes sense. Putting together unconditional and conditional estimates, Akogun and co. argue that that about 60 percent of the activity association with earnings is driven by showing up for work and 40 percent by more intense activity.
What were the measurement issues? It turns out that Nigerian sugarcane workers, just like me, lose their Fitbits. Of the 83 that start with the devices, 25 have lost them by the end of the study (and no observable variables predict the loss). This seems to be fairly common among the (few) studies that have tried to use these – Akogun and co. note that other studies have found up to 40 percent of individual-day data to be not usable. Akogun and co. close with the argument that further work needs to be done to see what can be done to increase compliance with measurement. They also raise a second issue/question: where is the optimal place to clip the FitBit. They note that it may be worth comparing the use of multiple accelerometers versus singles to tease this out.
And Malaria
The malaria results are interesting (as they were in the original paper). Akogun and co. have set up the experiment so that they can measure a) the simple impact of offering malaria treatment and testing (the ITT), b) the impact of being tested and found to be positive and treated (ToT), and c) the impact of being tested but not infected (treatment on the medically untreated). Now, the plantation shut down for a couple of weeks during their experiment, so they have some power issues, but with the sample they do have they find some changes in labor patterns. Overall testing leads to a shift from light to fairly active effort. Conditional on working, the effects are stronger, with a similar shift out of light activity to fairly or very active activity on the order of 0.8 hours per work day. This increases output by about 25 percent.
For those tested and found to be positive for malaria, the shift out of light activity (again conditional on working) is larger – about double the ITT estimates. This makes sense – when you get treated for malaria, you are more productive than folks who are sick and not treated. However, in line with their earlier paper, there is also an effect for folks who aren’t sick and find out the good news. Conditional on working, these folks are also experiencing a significant increase in fairly active hours, and this seems to be coming mostly from a decline in sedentary hours. So finding out you’re healthy seems to motivate folks to work harder.
This paper shows us a neat way to measure actual effort when we try and measure labor. By doing it in a setting where output and hours worked are cleanly measured, Akogun and co. have given us some clear confidence that this measure works. It clearly has some way to go on adherence and maybe precision, but it’s a big step forward.
Join the Conversation