Surveys on honey bee colony losses have been conducted by many researchers over the years to understand the factors that contribute to colony losses. Recognizing the importance of standard questionnaires for use in surveys, a network of honey bee specialists preceding the establishment of the COLOSS Action network, initiated by Cost Action FA0803, established at its first meeting a working group (Working Group 1 - WG1) whose aim was to develop and implement research surveys for the purpose of identifying such factors. The working group currently represents a global network of scientists who monitor colony losses. This group was conscious of the fragility of many survey results and addressed crucial issues to obtain a valid research framework (Van der Zee et al., 2012). Using other literature sources, the group developed and/or recognized appropriate case definitions, statistics and relevant factors associated with honey bee colony losses. The present manuscript aims to make the results of these efforts available to all researchers working in this field and to provide guidelines for conducting effective surveys.
Conditions in which to perform surveys on honey bee colony losses and achieve results which meet methodological standards are very different between and within countries. The present chapter offers guidelines to attain good quality surveys, even under unfavourable conditions. The main objective of these surveys (section 2.1.) is the estimation of winter colony loss rates, identification of specific areas with a higher or lower risk of honey bee mortality and information on possible determinants such as the control of Varroa destructor. This will enable the provision of advice on loss prevention and control.
We are conscious that the case definitions we present (section 2.3.) may be refined or changed in the following years because not enough knowledge is yet available to resolve many important issues. However what we present here does, in our view, give a good set of standards to which all researchers in this field should aim to conform in order to produce robust and reliable results.
The target population of the surveys is usually the set of active beekeepers in a country or specific area. The possibilities for reaching the target population vary between and within countries. Sometimes registers of beekeepers can be used for collecting data; but more often, cooperation with beekeepers’ associations is necessary (section 5.). In some situations, both are absent and the investigator has to develop other survey strategies. Suggestions are given for sampling frames in situations where cooperation with a beekeeper association is not possible or if a beekeeper infrastructure is absent (section 6.4.).
The sample selection method used is one of the main issues for obtaining reliable survey outcomes. Selecting a random sample of beekeepers gives results whose accuracy can be quantified (section 6.1.), if, as is usually the case, studying the whole target population instead of a sample is not feasible. At present most monitoring surveys with questionnaires will gain in quality if the shift is made from the present common practices of self-selected samples (samples in which participants volunteer to take part) towards at least simple random sampling. The same or better survey results may be achieved by using other more sophisticated forms of probability sampling, and relatively small sample sizes might then be sufficient.
Detailed consideration will also be given to the various sources of bias which may affect survey outcomes (De Leeuw et al., 2008), whose effect will usually be to introduce errors into results whose effects are difficult or impossible to assess. In particular, it is good practice to strive for high response rates, although there is no empirical support for the notion that low response rates necessarily produce estimates with high nonresponse bias (Groves, 2006). However that risk is inevitably present if response rates are low.
Attention is given here to a variety of methods of statistical data analysis. These range from simple analyses to examine the effects of different Varroa controls or other individual risk factors on mortality, to more advanced methods involving the use of Generalized Linear Models (GZLMs) to investigate simultaneously the possible effects of multiple different factors on colony loss rates. We use the statistical program R to illustrate an analysis in section 10., using data from the Netherlands. Survey design and sampling methodology is illustrated throughout the manuscript using Scotland as a case study. An introduction to this is provided in Box 1.