We test our hypothesis by relating migration, meteorological, and climate change literacy data for 37,000 individuals surveyed in 30 African countries between September 2016 and September 2018 to climate change occurring in the respondents’ home location measured at a 0.25*0.25° grid resolution.Footnote 2
As shown in Fig. 1, there is substantial within-country climate variation in the African countries we consider; this variation is linked to higher temperatures compared to earlier decades, consistent with the notion of global warming. We exploit this variation to estimate the effect of climate variability on individual migration intentions (for a further contextual overview on environmental change and migration in Africa, see, e.g., Morrissey 2014; Brüning and Piguet 2018; Borderon et al. 2019).
To estimate the effect of climate change and climate change literacy on migration intentions, we estimate the following model:
Using the OLS estimator, we explain migration intentions (MI) of individual i living in interview location r by a measure of climate variability (TEMP), usually an indicator of rising temperatures, climate change literacy (CLIMLIT), and their interaction. We also account for a set of controls (X) that include climate zone and year–month fixed effects and variables measuring individual characteristics of respondents (e.g., age, sex). Thus, our models compare migration intentions of individuals with different levels of climate change literacy within the same region after accounting for any country-specific and climate zone–specific effects and temporal trends.
We are primarily interested in the interaction effect of increasing temperatures and climate change literacy on migration intentions. This allows us to evaluate whether individuals with different climate change literacy levels respond differently to changing climate conditions. Summary statistics for all variables are reported in the Supplementary Tables S1a and S1b.
We employ a set of Climate Extreme Indices (CEIs), defined and developed by the Expert Team on Climate Change Detection and Indices (ETCCDI) and adopted by the World Meteorological Organization (WMO) (see Mistry, 2019; Zhang et al., 2011) for a full description of the CEIs). The calculation of CEIs at annual timescales requires daily measures of (i) maximum near-surface air temperature (Tmax, in °C), (ii) minimum near-surface air temperature (Tmin, in °C), and (iii) near-surface total precipitation (Precip, in mm). We rely on the Warm Spell Duration Index (WSDI) as our principal measure. It measures the annual number of days in which the daily maximum temperature (Tmax) exceeds the 90th percentile of the long-term average Tmax for at least six consecutive days.
The input meteorological variables are drawn from the ERA5 global reanalysis data product at 0.25° gridded resolution, made available by the European Center for Medium-Range Weather Forecasts (ECMWF) through the Copernicus Climate Data Store.Footnote 3 The CEIs are assembled using the WMO-recommended R software package “Climpact2.”Footnote 4 The CEIs employed in our study are further defined in Supplementary Table S2. In short, the CEIs in the econometric framework are operationalized as follows:
WSDI annual trend, 1979–2016: Trend in Warm Spell Duration Index, values 1979–2016
WSDI annual trend, 2009–2016: Trend in WSDI, values 2009–2016
WSDI change (2009–2016)-(1979–2008): Long-term difference in WSDI values, 2009–2016 average values minus 1979–2008 average values
WSDI 10-year average: 10-year average in WSDI, values 2007–2016
WSDI 3-year average: 3-year average in WSDI, values 2014–2016
HWM 10-year average: 10-year average in heat wave magnitude (Perkins and Alexander, 2013), values 2007–2016
SPEI 3-year average: 3-year average in Standardized Precipitation Evapotranspiration Index (Vicente-Serrano et al., 2010), values 2014–2016
Migration intention data
The data comes from the 7th round of the Afrobarometer (2019), in which face-to-face interviews with a randomly selected sample of 1200 to 2400 respondents in 30 African countries were carried out between 2016 and 2018. For our analysis, this results in a sample of over 37,000 interviewed individuals. Using geo-coded information (longitude and latitude) provided by the Afrobarometer, we link each interview location in the Afrobarometer to the corresponding local climate data.
We operationalize individuals’ migratory responses as their migration intentions. Migration intentions are measured by considering respondents’ answer to the following survey question: “How much, if at all, have you considered moving to another country to live?” Answer options are “Not at all” (coded 0), “A little bit,” “Somewhat,” and “A lot” (all coded 1); non-responses are dropped from the sample. Note that the survey specifically asks for international migration intentions; while domestic movements may also be a likely response to climate change, recent evidence suggests that the effect of environmental stress is particularly pronounced for the relative increase of intra-regional migratory behavior (Bekaert et al. 2021).
Climate literacy data
We consider someone as climate literate if the survey respondent has heard of climate change and indicated that climate change means negative changes in the weather (such as more droughts, floods, or extreme heat) and/or thinks that climate change makes life in their country somewhat or much worse. Persons who are climate illiterate have never heard of climate change or think that it leads to positive changes in the weather or that it makes life in the country better due to changes in environmental conditions (see Table S1c in the Supplementary Material for question wordings). On average, 44% of our respondents are climate literate (see Table S1a in the Supplementary Material and also Selormey et al., 2019: 8, 12). The share of climate literacy varies between circa 26% (Tunisia) and approximately 75% (Uganda).
As we show below, climate change literacy is highly correlated with general education. We, however, prefer to account for climate change literacy rather than education for three reasons. First, the education variable in the Afrobarometer dataset is a rather coarse measure of school degrees or length of school education that are difficult to compare across countries with different school systems. They also do not tell us anything about the content or the quality of education. Second, climate change literacy directly measures our theoretical construct, namely knowledge of climate change and its implications. While this knowledge might have been acquired in school, our measure helps us show that this knowledge is still available. Third, we may consider climate literacy as a means of correcting for a potential difference between meteorological reality and individual climate change perceptions. For instance, DeLongueville et al. (2020) show that changes in local rainfall are not necessarily reflected in rainfall perceptions among rural dwellers in Burkina Faso, arguing that subjective perceptions are influenced by additional (individual) factors (see also Niles and Mueller 2016).
However, we may expect climate literate individuals to be less subjective about climate change, making it easier for us to relate meteorological reality to the individual migratory response to climate change. Indeed, in Figure S1 in the Supplementary Material, we show that climate change literacy increases with rising temperature trends, where variation is substantial, especially in areas with higher WSDI levels. This supports the notion that climate literacy is linked to meteorological reality, while at the same time being—by construction—linked to individual climate change perception.
Data on the control variables also come from the Afrobarometer (2019). In the baseline model, we control for age and gender. Here, controlling for age is especially necessary because respondents’ age may correlate with both exposure to long-term climate change (e.g., older individuals are more likely to experience long-term changes) and migration intentions. We also consider a set of country fixed effects to account for economic and institutional differences between countries that may affect migration intentions as well as climate zone fixed effects to consider the differential role of increasing temperatures on human behavior in different climate zones. Here, we use the usual Köppen-Geiger climate classification system. Month–year fixed effects are included in some specifications to control for potential systematic differences in responses due to different interview periods.
Correlates of climate change literacy: model and variables
Climate literates are not randomly distributed. Thus, besides comparing whether individuals who are knowledgeable about climate change have a different migratory response to it, we also want to understand which individuals are well informed about it in the first place. Therefore, we run additional OLS models with various fixed effects to identify individual socio-economic correlates of climate change literacy. These correlates, in turn, are expected to deepen our understanding of the nexus between climate change literacy and migration intentions under climate change.
We identify the various correlates of climate change literacy by considering the following model:
Climate change literacy (CLIMLIT) is operationalized as explained above. The vector X includes the same fixed effects as defined for model 1. CORR refers to one of the following variables, which are all drawn from the Afrobarometer (2019):
News consumption: We differentiate between persons who get their news from any media (radio, television, newspapers, internet, or social media) every day, a few times a week, a few times a month, less than once a month, or never.
Profession: We consider individuals who are out of work (e.g., housewives, unemployed persons), farmers and agricultural workers, low-skilled service and industrial workers (traders, cleaners, domestic helps, laborers, etc.), students, and specialists (i.e., high-skilled service and industrial workers, senior managers, teachers, lawyers, bureaucrats).
Education: We differentiate between no education, primary education, secondary education, and post-secondary education.
Location: We differentiate between rural and urban location.