STATISTICAL-ANALYSIS OF COMPUTER-GENERATED MEASUREMENTS FROM MANUALLYOUTLINED POLLEN PERFORATIONS

Quantitative Pollen-sculpture variation can provide systematically use ful information. To test manual-outlining techniques used with automat ed measurement of pollen-sculpture features, we conducted several expe riments in which perforations were traced with felt-tip pen onto thin sheets of plastic. In each experiment, seven individuals made five tra cings of the same pattern. Tracings were then digitized and perforatio ns measured using image-analysis software. Three test patterns were us ed: (1) an SEM micrograph with ''real'' pollen perforations; (2) a com puter-generated test pattern that simulated very small pollen perforat ions; and (3) a computer-generated test pattern that simulated very la rge pollen perforations. Experiments examined the effect on precision and accuracy of using: (1) opaque-black versus transparent-red ink; (2 ) small versus large perforations; and (3) fuzzy- versus sharp-edged p erforations. Tracings were statistically compared using the coefficien t of variation and an analysis of variance. Tracings of computer-gener ated test patterns also were compared directly with the digitized test pattern. Results indicated that: (1) variation between tracings by a single individual will usually be very small; (2) variation between in dividuals usually will be statistically significant, even with the sha rpest possible perforations; (3) tracing with opaque ink is more accur ate than with transparent ink; (4) both precision and accuracy increas e with perforation size; and (5) deviation of a single-tracing measure ment from the parametric value usually will be much less than five per cent. These experiments demonstrated that manual tracing of pollen per forations prior to image analysis can produce repeatable measurements accurate to two significant figures.