|EBERT, TIMOTHY - University Of Florida|
|ROGERS, MICHAEL - University Of Florida|
Submitted to: Journal of Economic Entomology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/11/2018
Publication Date: 4/28/2018
Citation: Ebert, T., Backus, E.A., Rogers, M. 2018. Handling artificially terminated events in electropenetrography data. Journal of Economic Entomology. 111(4):1987-1990. https://doi.org/10.1093/jee/toy117.
Interpretive Summary: Proper statistical methods are crucial to validity of most conclusions drawn from research. Observation of insect pest feeding on crop plants often encounters the problem of premature (that is, artificial) termination of insect feeding events, because observation may end before the insect has naturally ceased feeding. Therefore, appropriate statistical methods must be developed to compensate for artificially terminated events, or conclusions drawn may be incorrect. This is especially important for studies using electrical penetration graph (EPG) monitoring or video-recording of insect feeding, because no further measurement is possible after recording has ceased. The present research used a small simulated data set for illustrative purposes. The improved statistical analysis methods resulting from this research will enhance use of EPG for development of crop varieties resistant to feeding of piercing-sucking insect pests, as well as other statistical studies of pest feeding.
Technical Abstract: Behavioral assays can have a final behavior of unknown duration due to termination of the experiment before the behavior has ended. One application is in the recording of hemipteran feeding behavior using electrical penetration graphs (EPG). Within the literature on hemipteran feeding, it has been argued that because successive events of one waveform type (e.g. phloem ingestion, or xylem ingestion) have longer durations than preceding events, artificially terminated long events should be retained in the data. This is because discarding such data would bias the estimate of the mean more than retaining the data. Computer simulation was used to more rationally examine what effect this decision has on interpreting the data, and the problems that arise when observations are not independent of one another. The presence of a pattern in the duration of successive observations violates assumptions of randomness present in most statistical tests. Thus, it was ultimately concluded that the best choice is to handle the data in such a way that there is no pattern and, under those conditions, to remove terminal event durations from further analyses. Additional discussion is provided for dealing with artificially terminated events when there is no pattern in successive events. This includes some options for retaining these events out of biological necessity and accepting the resulting problems as unavoidable.