Climate change is a problem that has evolved and will continue to evolve over a long time frame, with causal antecedents and consequences often separated by
substantial periods of time. Predictions about the effects of climate change on psychological variables will be of most value if they can specify when these effects occur. Such predictions require the use of theory and analyses that explicitly take into account the role of time. Psychologists studying climate change impacts will need to be aware that the effects of exposure to particular environmental conditions may differ substantially depending on the time of exposure. For example, sustained changes to mean temperatures may affect human behaviour in different ways than do random short- term changes in temperature. This was a point that was made in the empirical articles presented in this thesis.
One important way in which psychologists may study the effects of different timeframes of exposure to particular environmental conditions (e.g., warmer
temperatures) is by studying the relationships between human behaviour and different components of variation in environmental conditions. The idea that series of
observations can be separated into different components of variation is widely recognised in fields such as econometrics that make extensive use of time series analysis (e.g., P. S. Mann, 1994), but is less widely appreciated in psychology. The analysis of the effects of different components of variation in temperature was extensively used in the empirical articles in this thesis. Doing so facilitated an examination of whether brief exposure to warmer temperatures (as in the analyses of irregular daily temperature variation), had similar effects to more sustained exposure to warmer temperatures (as in the analyses of seasonal and geographical variation in temperature). Finding ways to study the effects of more sustained exposure to particular environmental conditions is important for psychologists interested in studying the impacts of climate change given the limited timeframe of observations that may be available. In discussing the study of the effects of different components of variation in climatic conditions, two components of variation deserve particular attention:
6.4.1Studying geographical variation in environment and behaviour. By studying the differences in behaviour between populations living in different geographical areas, psychologists can make tentative observations about the effects of sustained, long-term, differences in environmental conditions. There are, however, two important challenges that arise when studying the effects of geographical variation in environmental conditions: confounding, and small sample size.
The first problem with studying the effects of geographical variation in
environmental conditions is the vulnerability of such analyses to confounding. Human populations living in different areas will differ in terms of many social, economic, demographic and cultural variables that may impact the behaviours being studied (e.g., assaults, suicide, etc.). Many of these extraneous variables may have some statistical relationship with the environmental variables being studied as causal inputs. This is a difficult problem to rectify firstly because there are many such extraneous variables, making it difficult to identify, measure and statistically control for the most important potential confounds. Secondly, determining whether a given extraneous variable is truly a confounding variable requires knowledge about the causal effects of variables on one another rather than just the statistical relationships between them. For example, Hsiang et al. (2013) argue against controlling for socioeconomic variables when examining the effects of temperature on conflict, because it is difficult to exclude the possibility that temperature itself affects such socioeconomic variables—in which case they would not have the formal characteristics of confounding variables25. Statistically controlling for a variable that is correlated with the independent variable but also affected by the
independent variable (thus not having the characteristics of a confounding variable) can bias the estimate of the effect of the independent variable (McNamee, 2003; Weinberg, 1993). Researchers attempting to analyse the effects of geographical variation in environmental conditions (e.g., temperature) on psychological variables will therefore have to put a great deal of thought into which statistical controls to apply. In many cases this will mean not just examining correlations between potential confounding variables and the independent and dependent variables studied, but also examining the related literature to determine the likely causal mechanism of relationships found: That is,
25 A confounding variable is one that is correlated with the independent variable, not affected by
the independent variable, and that affects the dependent variable (see McNamee, 2003; Weinberg, 1993). In contrast, an extraneous variable is any variable other than the independent variable that affects the dependent variable.
based on the current state of knowledge about causal effects in the domain studied, which variables are most likely to have the characteristics of confounds?
The second challenge when examining the effects of geographical variation in environmental conditions is that of dealing with small sample sizes and corresponding large quantities of sampling error. Although the methods used may differ somewhat from study to study, in general a recording of behaviour within one geographical area (no matter how long the time period) represents a single observation in an analysis of the effects of geographical variation in environmental conditions on human behaviour. This meant, for example, that the studies of geographical variation in temperature and the incidence of violent behaviour presented in this thesis had sample sizes of just 66 or 67 districts (this being the number of districts in New Zealand, depending on whether or not the Chatham Islands were included). The resulting inferences were thus subject to a great deal more uncertainty than those produced in the analyses of irregular variation in temperature, in which the sample sizes involved were thousands of dates. Specifically, the confidence intervals for the effect of geographical variation in temperature on the three types of violent behaviour studied were very wide. In this case the sample of geographic regions studied (New Zealand) was not ideal, covering a relatively small geographical area itself divided into a reasonably small number of regions. The obvious solution to this problem would be to sample data from a wider geographical area, thus reducing sampling error in geographical analyses. Sampling from a wide geographical area also has the added benefit of allowing researchers to study the effects of a wider range of values in the environmental predictors of interest (e.g., a wider range of temperatures), thus avoiding any problems with restriction of range. It also makes it more likely that populations in areas particularly vulnerable to climate change (e.g., those that are economically, socially and politically marginalised; see IPCC, 2014b) will be studied. In comparing populations across large and diverse geographical regions, however, researchers will need to be particularly wary of the problem of confounding by social, economic, demographic and cultural variables (as discussed above). They will also need to attend carefully to practical problems such as differing data collection practices across different geographical areas (a problem that came up in Study Three; see section 4.3.2).
6.4.2Studying trends in environment and behaviour.
One other component of variation in environmental conditions also deserves some special consideration with respect to its relationship with human behaviour. This component of variation is that of long-term trends in environmental conditions. On the face of it, an ideal way to study the likely effects of sustained changes in climate on human behaviour would be to obtain long series of observations of environmental conditions and the behavioural variables of interest. One could then remove irregular short-term variation in each series (e.g., by aggregating data over years or decades), and then study how the long-term trends in these environmental and behavioural variable relate to one another. Such studies of long-term trends have occasionally been
conducted in the temperature-violence literature (e.g., Helama et al., 2013; Holopainen et al., 2013).
Unfortunately, this approach comes with several challenges. The first is simply the practical difficulty of obtaining long and consistently recorded series of observations of the variables studied. This task may be feasible for meteorological variables, but is often difficult for behavioural variables. The second challenge is the problem that the climatic stability of the Holocene epoch (Dansgaard et al., 1993) means that there may be relatively little variation in long term trends in meteorological variables to study when analysing data from the past. This is despite the fact that there may be a great deal of fluctuation in meteorological variables over short periods of time. In effect, this is a problem of restriction of range that impacts the capacity to discover relationships between climatic and behavioural variables.
Finally, when comparing human behaviour across long periods of time, there is again the possibility of confounding due to changing social, demographic, economic and cultural conditions—much as is the case for analyses of geographical variation. These problems of confounding can largely be circumvented when analysing the effects of short-term (e.g., daily) variations in environmental conditions by the simple
expedient of controlling for the effect of time period. For example, in many of the empirical analyses reported in this thesis, controls were applied for the year of observation, providing a flexible control for any potential confounds that varied
reasonably slowly over time. But when analysing long-term trends, it is not so simple to control for time-trending confounds in this simple fashion. If using series of annual data, for example, one may not be able to control for the year of observation while
retaining an over-identified model (depending somewhat on the method used to apply the control). This means that when analysing long-term trends, just as when analysing geographical variation in environmental conditions and behaviour, researchers will need to specifically measure and control for potential confounding variables. Of course, obtaining observations of potential confounding variables over long periods of time may itself present difficulties. The study of long-term trends in environmental and behaviour variables thus could have some benefits, but comes with some significant challenges.