Variability in West African Peasant Cocoa Farms; Time Studies on Agodi Cocoa Yields

A group of small farms near Ibadan provided data on the annual yields of 2,480 cocoa trees for 18 years. These were entered on punch cards.

The correlation coefficient r (p, i, q) was calculated for the mean yields of units for an earlier period of p years and a later period of q years separated by an interval of i years. The units taken were single trees, one-thirtieth-acre plots and whole farms.

For all units r (1, i, 1) tends to decrease as i increases; but for single trees when i = 0 (and to a lesser extent when i = 2 and 4) it lies below the fitted curve for odd values of i. This is evidence of biennial bearing.

For plots and whole farms r was calculated for every combination of p = 1 or 2, q = 1, 2, 4 or 8, and i up to the maximum possible for each (p, q) combination for all possible years. The values obtained indicate that:

(a) One year’s pre-recording lessens the error variance of the first experimental year on average in the ratio of 0-5 : 1.

(b) The decrease is less for subsequent experimental years, the ratio being about 0-9 : 1 for an 8-year interval.

(c) The error variance of the mean yields over periods immediately following one or two years’ pre-recording is, after adjustment for pre-recording, mainly independent of the length of the period.

(d) Gain in accuracy from two years’ pre-recording is greater than from one, the difference generally more than justifying the cost of the extra year’s recording. But the extra pre-recording which delays the start of an experiment for a further year is not likely to be worth while.