Analyses, interpretation, and pitfalls

The designs featured in this section lend themselves to a straightforward analysis of variance testing the effect of date of treatment on colony strength parameters at the end of the study (Table 4). Depending on the presence of higher-order replications such as blocks, the investigator should be alert to interactions between treatments and the blocking factor which, if present, prescribe that the investigator test main treatments separately by block. The number of surviving colonies by the end of the study (n) may differ across treatments, so it may be necessary to accommodate unequal sample sizes through use of harmonic means transformation or lsmeans. Treatment means are separated (α ≤ 0.05) by a conventional test such as Tukey’s or Student-Newman-Keuls. If the investigator procured whole colony mite populations (see step 10 of section along with more user-friendly relative measures such as varroa board counts (see section 4.2.4. ‘Natural mite fall’) and mites per 100 bees (see section ‘Dislodging mites from bees’), then it is desirable to test the rigor of the relative measures at predicting real mite populations through the use of regression analyses testing linear, quadratic, and cubic terms. Ultimately, the investigator would like to deliver to beekeeper clients a user-friendly relative measure that accurately predicts real colony mite populations.

This analysis will permit the investigator to compare end-of-season colony condition across the various treatment regimens (times of acaricide application). The damage threshold is determined retrospectively as the highest average colony mite density at time of treatment associated with colony condition significantly non-different from positive controls at season’s end. In one real example, the threshold was defined as conditions that prevailed when colonies were treated in August because August-treated colonies fared as well statistically at season’s end as colonies treated continuously (Delaplane and Hood, 1999). The mid-season samplings permit the investigator to describe mite populations, user-friendly relative mite measures, and colony strength parameters that prevailed at the time thresholds were achieved. The highest, rather than lowest, retrospective mite density is used because of the conservative emphasis of IPM on prolonging the interval between treatments as long as possible. A low or zero pest tolerance is rare, unnecessary in the varroa / Apis mellifera IPM system, and more commonly associated with cropping systems for which pest-induced cosmetic damage is a problem with consumers.

This analysis will likewise identify mite densities that are irrecoverably damaging, in other words, mite densities at which point in time treatment does not prevent comparative colony deterioration by the end of the study. In another real example, it was shown that mite densities that prevailed in October exceeded a recoverable level because at season’s end the October-treated colonies were in significantly worse condition than continuously-treated colonies (Delaplane and Hood, 1997).

Alternatively, Strange and Sheppard (2001) defined damage threshold as: 1. the mite levels corresponding to colony treatment groups at season’s end with weight of bees less than initial starting levels (0.92 kg in this case); and 2. colonies with < 1150 cm2 sealed brood – a number derived from regression analyses predicting the amount of brood that should be present in colonies with 0.92 kg bees. With these boundary conditions the authors were able to retrospectively identify legacy mite levels that were either tolerable or irrecoverably high.

One pitfall in the field studies described here is a confounding effect, inherent to the design, between season (time of treatment) and colony mite densities. Mite population growth is regulated by length of brood-rearing season, ratio of worker brood to drone brood, and number of brood cells (Fries et al., 1994) and tends to increase over the course of the active season. Delaplane and Hood (1997) pointed out this confounding issue when they said, “Thus, our treatment threshold is reliable for August colonies meeting the conditions described in [the table showing retrospective colony descriptions], but may not be reliable for August colonies with significantly different amounts of bees or brood.” A more highly-resolved field model would replicate each of these terms independently within month of treatment.

Another pitfall comes from the emerging realization that honey bee morbidity is not always the product of a simple linear process or one factor, but rather a web of interacting factors (vanEngelsdorp et al., 2009). More sophisticated damage thresholds are needed that can integrate more than one morbidity factor and account for their possible interactions.


Pros: allows the definition of damage thresholds as basis of IPM implementation.

Cons: high workload, tedious.