The purpose of this invited paper is to stimulate, at the micro, meso, and macro levels, debate on positive analytical approaches to agricultural productivity. The session focuses on Africa because it is the only developing region where crop output and yield growth is lagging seriously behind population growth. The present paper treats the micro (farmhousehold) level. Macro and meso analyses are useful for tracing productivity effects of policy and infrastructure change employing broad samples over long periods. Micro analyses tend to track smaller samples over shorter periods, but dig below the aggregate surface to discern and explain productivity differences over crops, zones, and farmer groups. Micro analysis is useful for policy and technology formulation and for infrastructure and commodity promotion strategies. Further, many aggregate studies are limited to only a few composite product categories because of lack of more detailed data on labor, land, and capital allocations over crops. To obtain finer distinctions among crops, one needs detailed farm management surveys. Since the spate of African farm management studies in the 1960s and 1970s, soils have rapidly degraded, access to land has become increasingly constrained, factor and credit markets have changed structurally, and nonfarm activity by farm households has apparently increased. These changes should affect productivity across farm types, suggesting the need to revive attention to farm-level analysis. Past farm-level productivity work in Africa, which largely used positive analysis, tended to stratify the sample exogenously based on farm characteristics, generally by farm size, use of animal traction, access to credit, use of new seed varieties, land tenure status, or income. We follow these earlier studies by stratifying the farm household data by a factor we expect to affect technical parameters, namely farm capital. In the Guinean zone of Burkina Faso in particular, we stratify according to whether the household owns animal traction equipment (foll wing Barrett et al., and Jaeger and Matlon). Besides adding to empirical evidence on the productivity effects of animal traction, we endogenously stratify the sample with a selectivity model, an approach too rarely taken (with a few recent exceptions such as Carter) in productivity research. We then estimate production functions, controlling for selectivity bias. Such an approach also allows us to test the (indirect) effect of nonfarm income on productivity through its effect on technology adoption as embodied in farm capital acquisition. Studies of nonfarm income effects on capital acquisition are not absent in the literature (e.g. Barrett et al. in eastern Burkina), but they have been somewhat neglected in farm-level work, and neglected even more in aggregate-level research because of the rarity of rural nonfarm income data.