Dataset and Dependent Variable

The Norwegian Election Studies are conducted by Statistics Norway every four years in line with the national electoral cycle. They are based on a random sample taken from the nationwide population register. The surveys consistently achieve response rates above 50 per cent and the resulting samples are representative of the Norwegian population aged 18–79 years. Each survey includes detailed background information about respondents (including year of birth and age at the time of the survey) and covers questions about the elections as well as individuals' political attitudes and policy preferences. Since 1989, a question on respondents' preferences for protecting the environment has been included, which we employ as the basis of our dependent variable. The formulation used in the 1989 and 1993 surveys was as follows:

In the 1997 survey, the answer scale was extended such that it now started at 0 rather than 1. In the 2001, 2005, 2009 and 2013 surveys, the order of the preference options was reversed. For consistency, we recoded the answers provided during the first three surveys (1989, 1993 and 1997) to match the response order of the survey format used from 2001 onwards.

Important for our purposes, half of the sample in each survey wave was interviewed again four years later. This creates high-quality rotating panels of approximately 500 to 700 respondents (see Table 1), which we exploit to study repeated observations of the same individuals. Note, however, that the small changes in the survey design documented above create concerns about the rotating panels around the 1997 survey. With the scale reversal in 2001, respondents included in both 1997 and 2001 faced 10 as both a negative and positive statement. Moreover, the scale's extension in 1997 implies that respondents in both 1993 and 1997 faced (slightly) different scales. Both issues might affect their answers over time, which is our key interest in the analysis. Hence, both the 1993–1997 and 1997–2001 rotating panels are excluded from our final sample, leaving us with four rotating panels spanning a 24-year period (see Table 1). Appendix Figure A1 shows that most respondents shifted preferences at least somewhat over these four-year periods (that is, less than 30 per cent show no change at all). Some even document very substantial changes (we return to this below). These changes over time within the same individual are the main dependent variable in our analysis.

Table 1.

Structure of the rotating panel

Note: the response order and coding on our main variable of interest changed slightly in 1997, which undermines the comparability of the responses within the 1993–1997 and 1997–2001 rotating panels. As each individual appears in two subsequent surveys, the number of individual respondents equals 2,476. Note also that 2013 is the latest available survey at the time of writing.

Empirical Analysis

Method: Identification of the Life-Cycle Effect

Our aim is to identify a life-cycle effect in environmental concerns independent of any cohort and time effects. As mentioned, the key problem is the linearly dependent relationship Age = Time – Cohort, which makes independent identification of these three elements extremely challenging (Bell and Jones Reference Bell and Jones2014). Our approach is to focus on the first difference of the environmental concerns function. Clearly, cohort effects are eliminated by first-differencing the data because the same individual remains in the same birth cohort over time. Taking out cohort effects is a price we are willing to pay for being able to identify life-cycle effects.

Although the slope of the first difference (that is, the second difference) can be empirically identified, the first difference itself can only be estimated up to a constant, because potential time-specific effects (see Appendix Figure A2) could still contaminate the estimate of the life-cycle effect. The basic issue is that when four years have passed everyone is exactly four years older. This implies an unclear assignment of age and time effects when including both in an empirical model, such that the age at which environmental concerns reach a peak (or trough) cannot generally be identified. We follow Cheng, Powdthavee and Oswald (Reference Cheng, Powdthavee and Oswald2017) and address this problem by running a first-stage regression with the within-person change in environmental concerns as the dependent variable, and a full set of time dummies as independent variables (whereby survey years represent the time effects). The residuals from this regression reflect individuals' detrended change in environmental concerns. Any time-specific effects are eliminated from these residuals as the individual-level changes are scaled down by the average change in the data within a period.Footnote ²

We then run a second-stage regression with the detrended change in environmental concerns as the dependent variable, and age as the independent variable. This regression reveals the effect of a change in age – over and above the average time and age effects within a period – on the change in environmental concerns independent of any cohort effects. With subscripts i and t referring to individuals and time, respectively, this estimation approach can be written as:

(1)$$\Delta \rm Pref_{\it it} = \beta _0 + \beta _{\it t} {\rm time}\,{\rm dummies} + \varepsilon _{it}$$

(2)$${\hat \varepsilon }_{it} = \alpha _0 + \alpha _1{\rm Ag}{\rm e}_{it}( + \alpha _2{\rm Control}{\rm s}_{it}) + \mu _{it}$$

where ΔPref_it equals within-person changes in environmental preferences, and time dummies is a full set of survey-year indicator variables. $ {\hat \varepsilon }_{it} $ is the residual from estimating Equation 1, and indicates detrended changes in environmental concerns.Footnote ³ Age_it is respondents' age at the time of the second survey. Importantly, α ₁ ≠ 0 reflects non-linear life-cycle effects on environmental concerns. In particular, a combination of α ₀ > 0 and α ₁ < 1 is consistent with a positive first difference at young ages and a negative first difference at old ages (that is, an inverted-U shape over the life cycle). Note that α ₁ = 0 need not imply the absence of life-cycle effects, only that non-linear effects do not find support in the data. Any potential linear life-cycle effects cannot be identified independently of time effects, and are taken out of the estimations by the detrending procedure.

Although control variables are not strictly necessary, we sometimes extend our specification with two time-changing variables that capture respondents' real income (in 100,000 NOK, base year 2013) and education level (lower than secondary, secondary, higher education). These have been extensively analyzed as potential determinants of environmental preferences by, among others, Gelissen (Reference Gelissen2007), Franzen and Meyer (Reference Franzen and Meyer2010), Pampel and Hunter (Reference Pampel and Hunter2012), Jorgenson and Givens (Reference Jorgenson and Givens2014), Lo (Reference Lo2014) and Johnson and Schwadel (Reference Johnson and Schwadel2019). Summary statistics for all relevant variables are included in Table 2.

Table 2.

Summary statistics

Note: Preference for Environmental Protection is coded from 0 (‘Environmental protection measures should not be taken so far as to affect our standard of living’) to 10 (‘More should be done for environmental protection, even if it means people have their standard of living reduced to a considerable extent, yourself included’). Change in Preference for Environmental Protection is the within-person change in environmental concerns across two survey waves. Income measures respondents' real income in NOK 100,000 (base year 2013), while Education is measured in three stages as lower than secondary, secondary or higher education. The data are derived from a fully balanced rotating panel including surveys from 1989–1993 and 2001–2013.

Main Findings

We start our analysis by briefly looking at the cross-sectional relationship between age and individual preferences for environmental protection in Figure 1. The figure shows a strong inverted-U-shaped environmental concern function, which reaches its maximum value at 42.82 years.

Figure 1.

Cross-sectional relationship between age and preferences for environmental protection

Note: the figure depicts the relation between respondents' age and their preferences for protecting the environment measured on an 11-point scale. Dots reflect average preferences in one-year age bins, while the dotted line is a simple quadratic function fitted through the underlying data (with 95 per cent confidence intervals). Data cover a fully balanced rotating panel including surveys from 1989–1993 and 2001–2013.

Figure 1 provides some initial, suggestive verification that older individuals place less emphasis on protecting the environment. Yet, this figure obviously conflates time, cohort and life-cycle effects. To better understand life-cycle effects independently of cohort and time effects, we estimate Equations 1 and 2 using the fully balanced rotating panel covering the periods 1989–1993 and 2001–2013. The key findings are summarized in Figure 2 (and Appendix Table A1).Footnote ⁴

Figure 2.

Gradient of change in preferences for environmental protection by age

Note: the vertical axis depicts the de-trended within-person change in respondents' preferences for protecting the environment (with these preferences measured on an 11-point scale), while the horizontal axis depicts respondents' age at the time of the second survey. Dots reflect averages in one-year age bins, while the dotted line is a linear function fitted through the underlying data (with 95 per cent confidence intervals). The dataset covers a fully balanced rotating panel including surveys from 1989–1993 and 2001–2013.

The downward sloping line in Figure 2 indicates a negative relationship between the (detrended) change in support for environmental protection and an individual's age. Further, we find the first difference to be positive at young ages and negative at old ages. As argued above, these results provide strong evidence of an inverted-U-shaped life-cycle effect in environmental concerns. In terms of effect size, remember that our first-difference model effectively considers the survey-to-survey rate of change in expressed preferences. As such, each dot in Figure 2 is the average change in support for environmental protection among all people in the sample of a specific age. Given the four-year gap between surveys, the predicted decline in environmental preferences for 64 year olds is 0.168 compared to when they were 60 years old.Footnote ⁵ Extrapolating, the preferences of 72 year olds are predicted to be 0.624 lower than those of 60 year olds. At a mean value for environmental preferences of 5.7, this can be considered a substantively meaningful decline. Referring back to Figure 1, it indicates that a substantial share of the decline in environmental preferences between 60 and 72 years of age may be due to life-cycle effects.

Our findings are important, as the observed life-cycle effects will at least partly mitigate any cohort effect towards greater environmental concerns (Inglehart Reference Inglehart1971; Inglehart Reference Inglehart1981; Inglehart Reference Inglehart1990; Inglehart and Abramson Reference Inglehart and Abramson1994). Especially in rapidly aging societies – such as most advanced economies – and the concomitant increase in the median voter's age, this may have a considerable impact on future environmental policies.Footnote ⁶