2.1. Literature search

We searched the Web of Science on the portal available through the Potsdam Institute for Climate Impact Research (PIK) for articles that investigated current or future effects of various climatic factors on amphibians and reptiles. The initial criteria for article selection can be seen from the search terms listed in the electronic supplementary material, table S1. This initial search gave us a list of 1818 articles. We also included studies (n = 123) that we found in the process of contacting authors and reading other articles. After initial screening, we still had to assess 539 articles in more detail. Only 104 of those articles fit our selection criteria and were included in the analysis. The selection criteria were

(1) Articles had to be published between 2005 and April 2015, because we wanted to focus on the most recently published studies. Studies that were conducted before 2005 have already been included in previous reviews [23,25,26].

(2) The study included climatic factors, i.e. any factor that is directly affected by climate change. This includes, for example, temperature, precipitation, number of dry days, water temperature and storms (for a full list of climatic variables, see electronic supplementary material, table S2).

(3) The study was based on data collected in the field and conditions were not experimentally altered.

(4) Data were collected over a period of at least 5 years [27,28], including studies whose data (i.e. distributional information) came from museum specimens. We also included studies whose data were collected in disjunctive time intervals that were at least 5 years apart, and studies that modelled future distributions based on current distribution data from atlas data or long-term studies.

(5) The study analysed observed patterns to determine current or future effects of climate change, including

(a) a potential change of a trait (e.g. breeding date, distribution or body size) over time;

(b) a potential correlation between the temporal change of a species’ trait with changes in some facet of climate (e.g. temperature and rainfall volume);

(c) a potential change of some facet of climate over the period of the study; and

(d) a before–after comparison that documents the impacts of severe weather events.

(6) The original studies had to report the statistics needed to run a systematic review (i.e. p-values that are needed to assess whether associations with climate were unlikely to have been observed by chance). See below for explanation for using p-values and not more detailed information such as effect sizes.

Studies that reported a statistically significant relationship between a climate variable and a trait (p < 0.05) were categorized as showing a climate change effect and those with no statistically significant relationships (p > 0.05) were categorized as not showing a climate impact. Ideally, we would have considered the statistical power of the reviewed studies to detect significant relationships. However, this was not possible, because few studies included such estimations of statistical power. We used p-values rather than the sizes of reported effects (i.e. statistical effect sizes) because of the diversity of responses in the reviewed studies (see electronic supplementary material, table S2). Lack of consistency in the types of analyses and results reported further made the use of effect sizes impossible. Preliminary attempts to obtain more detailed statistical data directly from authors were mostly unsuccessful and were therefore not extended to all authors. Thus, our analyses were constrained by the limitations of the available data to using ‘vote counting’ rather than using formal meta-analytic methods [29].