Using Data Analysis to Assess the Impacts of NaAsO2 and As2O3 on Bee Mortality Rates

Background/Question/Methods

The canary in the birdcage is the classic early warning indicator, but other species may be viable alternatives. In this study, we re-evaluate data from Fujii (1980) to study the sensitivity of honey bee colony populations to arsenic poisoning. We expand on the initial analysis, fitting non-linear models to the temporal effects of arsenic treatment and evaluating both the explanatory and predictive power of our statistical models.

Results/Conclusions

Using a simple linear regression with arsenic dosage as the independent variable and the number of dead bees as the dependent variable, we find an expected statistically-significant relationship (β=2582, p=0.000; R2=0.50). Including binaries for each treatment and control type (As2O3 and NaAsO2), we recover our result of arsenic dosage (β=2068, p=0.012) but do not find any statistically significant effects specific to treatment type (R2=0.55). Modelling the data as a panel with a time random effect yields the same qualitative result for arsenic dosage (β=561, p=0.000), though we do find a positive effect of the NaAsO2 treatment (β=554, p=0.000). Log-transforming the data did not improve model fit. Using the data of one treatment to calibrate the model and the other treatment to evaluate the predictive capability of the model, we find that a linear model fit to the As2O3 data does a better job at predicting NaAsO2-caused mortality than the reverse. Our results have implications for the use of species as indicators of environmental health, as well as on honey bee population decline.