POST STRATIFICATION CLARIFIES TREATMENT EFFECTS ON PEARL-MILLET GROWTH IN THE SAHEL

Spatial variation in the growth of pearl millet [Pennisetum glaucum (L .) R. Br.] over short distances is a problem in field experiments in t he Sahel, but the causes are still poorly understood. Data from a 3-yr experiment with millet were used to compare four data types for their usefulness for reducing variation not related to treatment: (i) soil chemical data, (ii) residuals of the first year's yield data, (iii) a traditional fertility classification system, and (iv) plant vigor scor es. The completely randomized experiment consisted of four factors com bined to 48 treatments, replicated twice. There were three levels of m illet crop residues (CR), two levels of broadcast P, and four genotype s; the fourth factor had two levels and varied over years. Whereas che mical analyses of the topsoil did not explain overall variation, resid uals of plant scores used as covariates led to a reduction in residual variation of 32% for straw and 51% for grain yield in 1991. Most sati sfactory, however, was the use of residuals of plant scores to classif y plots into two strata of relatively low and high inherent soil produ ctivity (a retrospective procedure called post stratification). In low -productivity plots, a CR application of 2000 kg ha(-1) (compared with 500 kg ha(-1)) increased millet straw yield by an average of 42% and grain yield by 48% for the first 2 yr. In contrast, under high product ivity, yields were barely influenced by treatments. The application of P, however, was equally effective in both productivity strata. The re sults show that vigor scores can be useful to clarify treatment effect s on millet growth. The different responses of crop residues and P in the two productivity strata also indicate that nonchemical parameters such as soil mechanical resistance may contribute to soil microvariabi lity in the Sahel.