An investigation into the effect of climatic, ambient temperature on societal-level income inequality
Dublin Core
Title
An investigation into the effect of climatic, ambient temperature on societal-level income inequality
Creator
Sophie Lund
Date
2017
Description
Previous research has revealed contradictory findings concerning the relationship between temperature and behaviour. Some studies have found a warmer-is-better effect; warmer temperatures are associated with enhanced interpersonal interactions, including pro-social behaviours. Whereas other studies have found a warmer-is-worse effect; warmer temperatures are associated with negative social behaviours such as conflict, societal instability, crime and aggressive behaviours. The present study investigated the relationship between climatic, ambient temperature and societal income inequality. Climatic temperatures and Gini ratios (a measure of income inequality) were sourced from online databases for 29 countries across a range of time periods that fell between 1961 and 2015. A panel linear model analysis revealed that climatic temperature had no direct effect, nor lagged effect on income inequality. Therefore, the findings are not congruent with the warmer-is-better literature or the warmer-is-worse literature. Despite the null effect, the present study provides a further data point towards the debate concerning the effect of temperature on behaviour.
Source
Firstly, the study required Gini ratios of disposable, equivilsed income. The Gini ratio is a measure of income inequality whereby a ratio of 1 reflects perfect inequality (i.e. one household receives all of the income) and a ratio of 0 is indicative of perfect equality (i.e. income is equally shared across households). The ratio was calculated from disposable income, which is income after the deduction of taxes and social security charges. Additionally the ratio was equivilised which means that the ratio was adjusted to account for different sizes and compositions of households. Secondly, the study required mean climatic temperatures in degrees celsius.
Procedure
Gini ratios for 29 countries belonging to the organisation for economic co-operation and development (OECD) were sourced from several online databases that had calculated the ratios. The countries and years used in the present analysis were somewhat dictated by the availability of Gini ratios online and as a result the OECD countries Australia, Chile, Israel, Japan, Korea and Mexico could not be included in the present analysis and the year ranges included fell between 1961-2015. See table 1 for the sources of Gini ratios, and the countries and years for which Gini ratios were available.
It is important to note that the surveys from which the Gini ratios were calculated were slightly different, for example, some had different definitions of a ‘household’. Additionally, not all of the sources provided the exact Gini ratio calculation used.
Table 1: Online sources from which Gini ratios were obtained from several countries across several, differing, time periods
Country
Time period
Source of Gini ratios
Austria (AUT)
1995-2001, 2003-2015
Eurostat, European Union Statistics on Income and Living Conditions (2017).
Belgium (BEL)
1995-2001, 2003-2015
See Austria.
Canada (CAN)
1976-2015
Statistics Canada (2017).
Czechoslovakia (CZE)
2001, 2005-2015
See Austria.
Denmark (DEN)
1987-2015
Statistics Denmark (2017).
Estonia (EST)
2000-2002, 2004-2015
See Austria.
Finland (FIN)
1987-2014
OECD Data (2017)
France (FRA)
1995-2002, 2004-2015
See Austria.
Germany (GER)
1984-2013
German Socio-economic Panel Study (2015)
Greece (GRE)
1995-2001, 2003-2015
See Austria.
Hungary (HUN)
2000-2002, 2005-2015
See Austria.
Iceland (ISL)
2004-2015
See Austria.
Ireland (IRL)
1995-2001, 2003-2015
See Austria.
Italy (ITA)
1995-2001, 2004-2015
See Austria.
Latvia (LVA)
2000, 2005-2015
See Austria.
Luxembourg (LUX)
1995-2001, 2003-2015
See Austria.
Netherlands (NED)
2000-2014
Netherlands Central Bureau of Statistics (2017)
New Zealand (NZL)
1984, 1988, 1990, 1992, 1994, 1996, 1998, 2001, 2004, 2007, 2009-2014
Perry (2016)
Norway (NOR)
1986-2015
Statistics Norway (2017).
Poland (POL)
2001, 2005-2015
See Austria.
Portugal (POR)
1995-2001, 2004-2015
See Austria.
Slovakia (SVK)
2005-2015
See Austria.
Slovenia (SVN)
2000-2002, 2005-2015
See Austria.
Spain (ESP)
1995-2002, 2004-2015
See Austria.
Sweden (SWE)
1975, 1978-2013
Statistics Sweden (2017).
Switzerland (SWI)
2007-2015
See Austria.
Turkey (TUR)
2002, 2006-2013
See Austria.
United Kingdom (UK)
1961-2014
Institute for fiscal studies (2016)
United States (USA)
1967-2013
Proctor, Semega & Kollar, M. A. (2016).
Temperatures were sourced from the Climate Change and Knowledge Portal (2017) which contained the mean temperatures in degrees celsius for every country that was included in the present analysis for each month from years 1901-2015. Because we obtained mean Gini ratios for each year, we calculated mean climatic temperatures by calculating the average of the 12 months for each year, and country, that a Gini ratio was obtained. All Gini ratios and temperatures were accessed on 28th June 2017.
Design and analysis
In the present study the predictor variable was temperature and the outcome variable was Gini ratios. Data was collected for 29 countries across differing time periods ranging from 8-53 years resulting in a dataset with 594 observations. The dataset was a panel dataset whereby the data was cross-sectional (i.e. across countries) and longitudinal (i.e. across time periods) and unbalanced because of the differing time periods for each country. Therefore, to analyse the effect of temperature on Gini ratios, the plm package (Croissant & Millo, 2008) in R (R development core team, 2012) was used because this analysis has been designed to account for panel, unbalanced datasets. Additionally this package could determine whether country and time had an effect on Gini ratios and how these effects should be accounted for. The general linear model for the data set was (Croissant & Millo, 2008):
yit = α + Txit + µi + t + it
i = country
t = time
yit = Gini ratios
α = intercept
Txit the coefficient of the effect of temperature on Gini ratios
µi = the unobserved error as a result of the effect of country on Gini ratios
t = the unobserved error as a result of the effect of time on Gini ratios
it = residual/idiosyncratic error, independent of the predictor and individual error components
The specific model that was used in the present analysis was dependent on the existence of country effects (i.e. µi) and time effects (i.e.t) and the nature of these effects. There are three potential ways to model the panel datasets when estimating the effect of temperature on Gini ratios (Croissant & Millo, 2008):
1 – Pooled model; where time and country have no effect on Gini ratios (i.e. µi =0, t =0). Thus, the pooled models estimation is consistent and efficient, and applies across countries and time.
2 – Fixed effects model; where there are effects of country and/or time on Gini ratios and these effect(s) are correlated with the predictor variable, temperature. These correlated effect(s) result in the pooled models’ estimation being inconsistent because the estimates differ across countries and/or across time. Therefore, the fixed effects model accounts for the heterogeneity between countries and/or time by treating country and/or time as parameters to be estimated in the model and consequently the model gives consistent estimates of the effect of temperature on Gini ratios. This model can be one-way (i.e. the effect of country or time are taken into account) or two-way (i.e. the effects of country and time are taken into account).
3 – Random effects model; where there are effects of country and/or time on Gini ratios and these effect(s) are uncorrelated with the predictor variable, temperature. As a consequence of these uncorrelated effects, although the pooled models estimation is consistent, this estimation is inefficient. Thus, the random effects model accounts for the heterogeneity between countries and/or time by treating country and/or time as a separate error component(s) in the model and consequently the model gives consistent and efficient estimates of the effect of temperature on Gini ratios. Similar to the fixed effects model, the random effects model can be one-way or two-way.
To determine which model was appropriate for the dataset, and thus to determine the nature of the effects of country and time, we performed exploratory (i.e. graphical representations) and confirmatory (i.e. hypothesis testing) analyses. Firstly, we used graphs to visualise whether the intercepts were heterogeneous across countries and time as heterogeneity would suggest that the pooled model (i.e. model 1) was not appropriate. Following this, to determine whether the pooled model was appropriate for the dataset, we used the F test of stability, which by default tests whether the same coefficients applied to each country. Following this F test there were two potential routes.
(i) If the analysis revealed that the same coefficients applied across countries, we would then implement an F test of stability to test whether the same coefficients applied across time. If the second F test revealed that the same coefficients applied across time, a pooled model (i.e. model 1) would be used as this would provide a consistent and efficient estimation. Whereas if the second F test revealed that the coefficients did not apply across time, this would suggest that the pooled model was not appropriate and thus a Hausman test would be required to determine whether time should be modelled as a fixed effect (i.e. one-way fixed effects model; model 2) or random effect (i.e. a one-way random effects model; model 3).
(ii) If the analysis revealed that coefficients did not apply across countries this would suggest that a pooled model (i.e. model 1) was inappropriate for the dataset. Consequently a langrage multiplier test would be required to determine whether a one-way or two-way effects model should be used i.e. whether country alone had an effect on Gini ratios (i.e. one-way) or whether country and time had independent significant effects on Gini ratios (i.e. two-way). Secondly, a Hausman test would be necessary to determine whether the effect(s) should be modelled as fixed (i.e. fixed effects model; model 2) or random (i.e. random effects model; model 3).
Once a model had been specified, we estimated the direct and lagged effect of temperature on Gini ratios. Finally, we carried out diagnostic testing to analyse whether there was serial correlation or cross-sectional independence in the idiosyncratic errors of the model that would need to be dealt with.
Procedure
Gini ratios for 29 countries belonging to the organisation for economic co-operation and development (OECD) were sourced from several online databases that had calculated the ratios. The countries and years used in the present analysis were somewhat dictated by the availability of Gini ratios online and as a result the OECD countries Australia, Chile, Israel, Japan, Korea and Mexico could not be included in the present analysis and the year ranges included fell between 1961-2015. See table 1 for the sources of Gini ratios, and the countries and years for which Gini ratios were available.
It is important to note that the surveys from which the Gini ratios were calculated were slightly different, for example, some had different definitions of a ‘household’. Additionally, not all of the sources provided the exact Gini ratio calculation used.
Table 1: Online sources from which Gini ratios were obtained from several countries across several, differing, time periods
Country
Time period
Source of Gini ratios
Austria (AUT)
1995-2001, 2003-2015
Eurostat, European Union Statistics on Income and Living Conditions (2017).
Belgium (BEL)
1995-2001, 2003-2015
See Austria.
Canada (CAN)
1976-2015
Statistics Canada (2017).
Czechoslovakia (CZE)
2001, 2005-2015
See Austria.
Denmark (DEN)
1987-2015
Statistics Denmark (2017).
Estonia (EST)
2000-2002, 2004-2015
See Austria.
Finland (FIN)
1987-2014
OECD Data (2017)
France (FRA)
1995-2002, 2004-2015
See Austria.
Germany (GER)
1984-2013
German Socio-economic Panel Study (2015)
Greece (GRE)
1995-2001, 2003-2015
See Austria.
Hungary (HUN)
2000-2002, 2005-2015
See Austria.
Iceland (ISL)
2004-2015
See Austria.
Ireland (IRL)
1995-2001, 2003-2015
See Austria.
Italy (ITA)
1995-2001, 2004-2015
See Austria.
Latvia (LVA)
2000, 2005-2015
See Austria.
Luxembourg (LUX)
1995-2001, 2003-2015
See Austria.
Netherlands (NED)
2000-2014
Netherlands Central Bureau of Statistics (2017)
New Zealand (NZL)
1984, 1988, 1990, 1992, 1994, 1996, 1998, 2001, 2004, 2007, 2009-2014
Perry (2016)
Norway (NOR)
1986-2015
Statistics Norway (2017).
Poland (POL)
2001, 2005-2015
See Austria.
Portugal (POR)
1995-2001, 2004-2015
See Austria.
Slovakia (SVK)
2005-2015
See Austria.
Slovenia (SVN)
2000-2002, 2005-2015
See Austria.
Spain (ESP)
1995-2002, 2004-2015
See Austria.
Sweden (SWE)
1975, 1978-2013
Statistics Sweden (2017).
Switzerland (SWI)
2007-2015
See Austria.
Turkey (TUR)
2002, 2006-2013
See Austria.
United Kingdom (UK)
1961-2014
Institute for fiscal studies (2016)
United States (USA)
1967-2013
Proctor, Semega & Kollar, M. A. (2016).
Temperatures were sourced from the Climate Change and Knowledge Portal (2017) which contained the mean temperatures in degrees celsius for every country that was included in the present analysis for each month from years 1901-2015. Because we obtained mean Gini ratios for each year, we calculated mean climatic temperatures by calculating the average of the 12 months for each year, and country, that a Gini ratio was obtained. All Gini ratios and temperatures were accessed on 28th June 2017.
Design and analysis
In the present study the predictor variable was temperature and the outcome variable was Gini ratios. Data was collected for 29 countries across differing time periods ranging from 8-53 years resulting in a dataset with 594 observations. The dataset was a panel dataset whereby the data was cross-sectional (i.e. across countries) and longitudinal (i.e. across time periods) and unbalanced because of the differing time periods for each country. Therefore, to analyse the effect of temperature on Gini ratios, the plm package (Croissant & Millo, 2008) in R (R development core team, 2012) was used because this analysis has been designed to account for panel, unbalanced datasets. Additionally this package could determine whether country and time had an effect on Gini ratios and how these effects should be accounted for. The general linear model for the data set was (Croissant & Millo, 2008):
yit = α + Txit + µi + t + it
i = country
t = time
yit = Gini ratios
α = intercept
Txit the coefficient of the effect of temperature on Gini ratios
µi = the unobserved error as a result of the effect of country on Gini ratios
t = the unobserved error as a result of the effect of time on Gini ratios
it = residual/idiosyncratic error, independent of the predictor and individual error components
The specific model that was used in the present analysis was dependent on the existence of country effects (i.e. µi) and time effects (i.e.t) and the nature of these effects. There are three potential ways to model the panel datasets when estimating the effect of temperature on Gini ratios (Croissant & Millo, 2008):
1 – Pooled model; where time and country have no effect on Gini ratios (i.e. µi =0, t =0). Thus, the pooled models estimation is consistent and efficient, and applies across countries and time.
2 – Fixed effects model; where there are effects of country and/or time on Gini ratios and these effect(s) are correlated with the predictor variable, temperature. These correlated effect(s) result in the pooled models’ estimation being inconsistent because the estimates differ across countries and/or across time. Therefore, the fixed effects model accounts for the heterogeneity between countries and/or time by treating country and/or time as parameters to be estimated in the model and consequently the model gives consistent estimates of the effect of temperature on Gini ratios. This model can be one-way (i.e. the effect of country or time are taken into account) or two-way (i.e. the effects of country and time are taken into account).
3 – Random effects model; where there are effects of country and/or time on Gini ratios and these effect(s) are uncorrelated with the predictor variable, temperature. As a consequence of these uncorrelated effects, although the pooled models estimation is consistent, this estimation is inefficient. Thus, the random effects model accounts for the heterogeneity between countries and/or time by treating country and/or time as a separate error component(s) in the model and consequently the model gives consistent and efficient estimates of the effect of temperature on Gini ratios. Similar to the fixed effects model, the random effects model can be one-way or two-way.
To determine which model was appropriate for the dataset, and thus to determine the nature of the effects of country and time, we performed exploratory (i.e. graphical representations) and confirmatory (i.e. hypothesis testing) analyses. Firstly, we used graphs to visualise whether the intercepts were heterogeneous across countries and time as heterogeneity would suggest that the pooled model (i.e. model 1) was not appropriate. Following this, to determine whether the pooled model was appropriate for the dataset, we used the F test of stability, which by default tests whether the same coefficients applied to each country. Following this F test there were two potential routes.
(i) If the analysis revealed that the same coefficients applied across countries, we would then implement an F test of stability to test whether the same coefficients applied across time. If the second F test revealed that the same coefficients applied across time, a pooled model (i.e. model 1) would be used as this would provide a consistent and efficient estimation. Whereas if the second F test revealed that the coefficients did not apply across time, this would suggest that the pooled model was not appropriate and thus a Hausman test would be required to determine whether time should be modelled as a fixed effect (i.e. one-way fixed effects model; model 2) or random effect (i.e. a one-way random effects model; model 3).
(ii) If the analysis revealed that coefficients did not apply across countries this would suggest that a pooled model (i.e. model 1) was inappropriate for the dataset. Consequently a langrage multiplier test would be required to determine whether a one-way or two-way effects model should be used i.e. whether country alone had an effect on Gini ratios (i.e. one-way) or whether country and time had independent significant effects on Gini ratios (i.e. two-way). Secondly, a Hausman test would be necessary to determine whether the effect(s) should be modelled as fixed (i.e. fixed effects model; model 2) or random (i.e. random effects model; model 3).
Once a model had been specified, we estimated the direct and lagged effect of temperature on Gini ratios. Finally, we carried out diagnostic testing to analyse whether there was serial correlation or cross-sectional independence in the idiosyncratic errors of the model that would need to be dealt with.
Publisher
Lancaster University
Format
data/excel.xlsx
Identifier
Lund2017
Contributor
John Towse
Rights
Open
Language
English
Type
The Gini ratio
Coverage
LA1 4YF
LUSTRE
Supervisor
Louse Connell
Project Level
MSc
Topic
Cognitive Psychology
Sample Size
N/A
Statistical Analysis Type
regression- panel linear, two-way fixed effects
serial correlation
Bruesch-Godfrey/Wooldridge test
serial correlation
Bruesch-Godfrey/Wooldridge test
Files
Collection
Citation
Sophie Lund, “An investigation into the effect of climatic, ambient temperature on societal-level income inequality
,” LUSTRE, accessed April 20, 2024, https://www.johnntowse.com/LUSTRE/items/show/28.
,” LUSTRE, accessed April 20, 2024, https://www.johnntowse.com/LUSTRE/items/show/28.