Not Your Average Job: Measuring Farm Labor in Tanzania

Abstract

Understanding the constraints to agricultural growth in Africa relies on the accurate measurement of smallholder labor. Yet, serious weaknesses in these statistics persist. The extent of bias in smallholder labor data is examined by conducting a randomized survey experiment among farming households in rural Tanzania. Agricultural labor estimates obtained through weekly surveys are compared with the results of reporting in a single end-of-season recall survey. The findings show strong evidence of recall bias: people in traditional recall-style modules reported working up to four times as many hours per person-plot relative to those reporting labor on a weekly basis. Recall bias manifests both in the intensive and extensive margins of labor reporting: while hours are over-reported in recall, the number of people and plots active in agricultural work are under-reported. The evidence suggests that this recall bias is driven not only by failures in memory, but also by the mental burdens of reporting on highly variable agricultural work patterns to provide a typical estimate. All things equal, studies suffering from this bias would understate agricultural labor productivity. JEL Codes: C8, O12, Q12

Correlated non-classical measurement errors, ‘Second best’ policy inference, and the inverse size-productivity relationship in agriculture

“…This type of misreporting can vary systematically with the true value of farm size, since those with little land or output have little or nothing to hide. Second, farmers may not accurately recall information related to much earlier events; in particular, extended recall periods may cause them to forget details of past events (Beegle et al (2012); Arthi et al (2018)) or seasonspecific harvests (Ali et al (2009);Howard et al (1995)). 8 Third, precise and universally applied measurement units may not be widely employed in low-income rural areas where imprecise local measures are commonplace.…”

“…1 Of particular concern is non-classical measurement error (NCME), which occurs when the error in measuring a variable of interest is correlated with the true value of that variable, with the true values of other variables in the model, or with the errors in measuring those values (Bound et al (2001)). 2 Many papers have clearly demonstrated the widespread prevalence of NCME and its relevance for policy inference in a range of fields, especially labor (e.g., Borjas (1980); Bound and Krueger (1991); Bound et al (1994); French (2004); Kim and Solon (2005); Arthi et al (2018)), consumer behavior (Gibson and Kim (2010); Gibson et al (2015)), development (Baird andÖzler (2012); Beegle et al (2012); Chao et al (2012); Desiere and Jolliffe (2018)), health (Das et al (2012); Larsen et al (2017)), and agriculture (De Groote and Traoré (2005); Carletto et al (2013Carletto et al ( , 2015; Gourlay et al (2017)). The sensible guidance provided by that literature is to employ better measurement methods so as to reduce error.…”

“…Such recall bias affects many other agricultural metrics, including labor use(Arthi et al (2018)). 9 While most of the above reasons apply to farm size and production measurements, there are additional problems that may affect measurement of production.…”

“…For example,Arthi et al (2018) show that aggregating households labor use involves competing biases, over-reporting at the extensive margin of labor use and under-reporting at the intensive margin, with these errors ultimately cancelling each other out to minimize aggregate bias.…”

Assessing recall bias and measurement error in high-frequency social data collection for human-environment research

“…Our sample was purposively constructed to capture likely early adopters of smartphone technologies in this part of rural northern Bangladesh and skewed toward younger, more educated, and more male participants than a representative sample. Arthi et al (2018) recently found recall bias to be lower among the more educated in a high-frequency telephone survey experiment in Tanzania. To the extent that such results are generalizable to rural contexts like our Rangpur sample, this suggests that differences in recall identified in our study might be even further pronounced (and the method of smartphone-based data collection even more worthwhile to develop) for the broader population.…”