Table 1 below provides the results of both the Stage 1 and Stage 2 estimates of our IV model as well as the OLS estimates for comparison. In what follows, we present the coefficients of only those variables that are central to our analysis. Table 8 in the Appendix provides the summary statistics for all the variables used in our analysis and Table 9 provides the full set of results, including for the controls.
Starting with the Stage 1 results, we see that both the DG indicators (Delay 1 and Delay 8) that we use as instruments are positive and highly significant both for F&V consumption and for sports activity. Both of these delay variables have a larger impact on sports activity than on F&V consumption. The coefficient of the female variable in our Stage 1 results indicates that women eat more F&V than men and do less sports activity. Additionally, both instruments—Delay 1 and Delay 8—are positive and highly significant in determining F&V consumption. More precisely, Delay 8 (eating healthy because it pays off in the long run) has a larger impact than Delay 1 (the ability to stick to a diet) confirming that it is the ability to think of the long run that is crucial.
Turning to the Stage 2 estimates, we see first from the endogeneity test that with 99% confidence we can reject the hypothesis that our life-style variables are exogenous, thus justifying the use of instrumental variables in the models. We present here the results of a version of the Durbin-Wu-Hausman test valid for robust standard errors (Hayashi, 2000, pp. 233–34). We see that the Chi-squared value is very large and the P-value is very small indicating that our lifestyle variables are endogenous and need instrumentation. Our instruments also pass the over-identification Hansen J Test. The test statistic is not significant suggesting that we can reject the null of over-identification. The F-test value is very high, substantially above any typical threshold, confirming that both instruments are ‘good’ instruments.
As a robustness check we run our estimates using Delay 1 alone and Delay 8 by itself as instruments, and our results are similar, which provides further support that our instruments are valid. The results for these are provided in Table 10 in the Appendix.
We see a few more patterns. First, all the coefficients (both OLS and IV) are highly significant and positive indicating that both F&V and sports have a positive relationship with life satisfaction. Second, we see that the OLS coefficients are all downward biased and the IV estimates are higher than the OLS coefficients. Our results therefore lead us to conclude that both F&V consumption and exercise have a significant positive impact on life satisfaction.
Heterogeneity of Impacts
Our results so far clearly indicate that delaying gratification is a significant instrument for F&V consumption as well as sports activity (Stage 1). They also indicate that F&V consumption and sports activity are significant in influencing life satisfaction for both men and women (Stage 2). In this section, we will consider whether these results remain consistent across various characteristics. In particular, it is possible that the impact of lifestyle on life satisfaction might be different across gender, education, income quartiles, age groups, in rural vs urban areas and so on. We will explore some of these relationships in order to establish the robustness of our results.
The summary statistics in Table 2 below indicate that F&V consumption and sports activity increase monotonically with income quartiles as do the ability to delay gratification and the sense of control. There is no causality implicit here, only the association can be noted. The table also indicates that consumption increases by age group but, not surprisingly, sports activity decreases. The patterns of DG and sense of control are not clear. It is worth noting that life satisfaction is increasing monotonically across income groups; is higher in rural areas and lowest in the 25–54 age group.
The summary statistics for our heterogeneity variables are presented in Table 2. All values are statistically significantly different by characteristic except for life satisfaction scores for men and women. On a scale of 1–7, the average for men is 4.99 and for women is 5.02. Turning to the consumption of F&V, we find that men consumed 3.58 portions of F&V, while women consumed 4.07 portions. Thus, women consume significantly more F&V than men but men do significantly more sports than women with a score of 3.96 against 3.26 for women. Finally, the table indicates that women have significantly better ability to delay gratification when considering the longer term payoff of healthy eating (delay 8) than men do. While this is also true for the ability to stick to a diet (Delay 1), the difference here is much smaller. Overall, therefore, our summary statistics indicate that women have better consumption habits and give greater weight to the long run when making these decisions.
Turning to differences across income quartiles, we find that both F&V consumption and sports activity differ across income quartiles, as does life satisfaction. These differences are all significant. The ability to delay gratification (Delay 1) is significantly different between quartile 1 and quartile 4 (with people in the highest income quartile being able to stick to a diet significantly better than in the lowest) but the difference is not significant between quartile 1 and 2. Delay 8 is not significantly different between quartiles 2 and 3 but is significantly different across all other quartiles. People in the highest income quartile appear to be able to delay gratification better than people in the lowest. These differences are confirmed in the above/below median sub-samples.
Table 2 also indicates that F&V consumption and sport activity do vary by education. In particular, those with a degree consume more F&V and also do more exercise than those without a degree. The summary statistics also indicate that they are significantly more likely to say they have control over their lives and able to delay gratification than those without degrees. While rural areas consume more F&V and do more sports, there is no significant difference in their sense of control or their ability to delay gratification from urban areas. Finally, the differences in life satisfaction, lifestyle and control as well as delayed gratification are significant across all age groups that we consider (15–24, 25–54 and 55–64 year olds) (Table 3).
In what follows, we consider whether there is heterogeneity of our results across sub-samples. Considering first the differences across gender, we confirm, once again, that both F&V consumption and sports activity have a positive significant impact on life satisfaction. The diagnostic statistics indicate that lifestyle and life-satisfaction are endogenous and therefore instrumentation was necessary. The instruments pass the overidentification test and the F-test value is above the recommended threshold. However, there appears to be no significant difference in the impact that these variables have on life-satisfaction by gender. This is clear from the insignificance of the coefficient of the interaction term between gender and the two lifestyle variables and confirms the patterns indicated by our summary statistics.
Table 11 in the Appendix presents results separately by gender. These results confirm that even though both life-style dimensions have a positive and significant impact on life-satisfaction, their impact by gender differs only marginally in our sample.
We estimate the model to consider heterogeneity across the income quartiles in 3 different ways—by estimating across the 4 income quartiles, estimating for the above and below median groups and estimating the top quartiles vs the others. All three give us very similar results and we will therefore present the results for the top quartile vs others in this section, as it makes the pattern very clear. Once again, the endogeneity tests shows that we can reject with high confidence that lifestyle and life-satisfaction are exogeneous. They also confirm that our instruments are strong (Table 4).
Our results indicate, as before, that F&V consumption and sports activity have a positive, significant impact on life satisfaction. In addition, we see that being in the top quartile increases life satisfaction significantly. However, our results indicate that the interaction term between F&V and the top income quartile is not significant, leading us to conclude that being in the top quartile neither increases nor decreases the impact of F&V on life-satisfaction. Surprisingly, our results indicate that being in the top quartile and doing sports activity decreases life satisfaction. It is important to interpret this result carefully. Even in this group (top income quartile and doing sports), the impact of sports activity on life satisfaction is positive. However, the size of the impact is smaller than the impact of sports activity in the lower income quartiles. While this might seem surprising at first glance, it could arise from the fact that individuals in the top quartile might get a sense of achievement from their income rather than sporting prowess. Second, for this group, the need to do sports might decrease the time they have available for other things including their work and this therefore decreases their sense of satisfaction from doing sports. Despite this, it is worth reiterating that even in this group, sports activity has a positive impact on life satisfaction. Thus, our broad result is robust across income quartiles but we note that the size of the impact varies across income quartiles.
In Table 12 in the Appendix, we present the results separately for the top quartile and for all quartiles below the top one. While we cannot see from this table if the differences are significant across income categories, we can see that the size of the impact of both F&V and sports is larger in the ‘below top income quartile’ than in the ‘top income quartile’ confirming the results above.
Our IV results (Table 5) indicate that, having a degree does not significantly influence life satisfaction. Once again, our diagnostics indicate the need for instrumentation and also that our instruments pass the over-identification test and the F-test. Additionally, amongst those with degrees, consuming more F&V does not lead to higher life satisfaction. However, this is not true for sports activity. Again, it is worth emphasising that sports activity increases life satisfaction (with a coefficient of +0.191), even amongst those with degrees but it has a smaller positive effect for degree holders (+0.191–0.081) than for those without degrees. Again, this could be because those with a degree derive a higher life-satisfaction from achievements other than sports and because sports reduces their time for other activities.
This can be clearly seen from Table 13 in the Appendix where the results are presented separately for those with a degree and those without. The results show that while both lifestyle variables have a positive and significant impact on life-satisfaction for those with and those without a degree, the impact is larger in the ‘No degree’ category. Again, the results in Table 13 corroborate the ones in Table 5 above.
Even though both F&V and Sports Activity have a positive and significant impact on life-satisfaction, and even though our summary statistics indicated that those living in rural areas eat more F&V and do more exercise, our results in Table 6 reveal that there are no significant urban/rural differences in terms of the effect of lifestyle on life-satisfaction. The coefficients of the interaction terms between lifestyle and rural/urban are not significant. These results are reinforced by the ones in Table 14 in the Appendix, where we present the results separately by regional division. The results in Table 14 show that even though the coefficients of F&V and Sports Activity are larger in the rural area, the difference is only marginal. Hence the results are robust across this regional division as well.
Finally, Table 7 provides the results across three age categories: 16–24, 25–54 and 55–64 year olds. In these estimations, the instruments do not pass the over-identification test. However, we present the results for comparison purposes. These results indicate that the effect of lifestyle reduces with age (which only shows up in the IV estimates).Footnote 4 There is significant heterogeneity across age groups. To start with, it seems clear that the young (16–24 year olds) and the old (55–64) have higher life satisfaction than the middle aged. Interacting age group with F&V consumption, we find that there are significant negative coefficients for both these age groups indicating that amongst the 16–24 year olds, life satisfaction is lower if they have to eat F&V compared to the middle age groups (a coefficient of − 0.185). This is also true for the 55–64 year olds (− 0.146). It is important to emphasise that, despite this, the coefficient for this group is positive i.e. that F&V consumption does increase life satisfaction (0.746–0.146) even for this group [though the increase would be smaller than for the 25–54 year olds (1.208–0.185)].
The same pattern is clear for sports activity, where once again, the age group * sports activity coefficients are negative and significant. They lead us to conclude that within the 16–24 year olds, doing sports activity decreases life satisfaction below those who do not (− 0.096), even though the impact of sports activity remains positive and significant even for this group. This is also true for the 55–64 year olds and their discount is larger than for the younger age group (− 0.116). Even for them, however, the impact of sports remains positive. Thus, we can conclude that our broad result—that F&V and sports activity have a positive impact on life satisfaction—remains valid across age groups. However, the size of the coefficient varies across these groups, with the largest impact being amongst the middle aged (25–54 year olds).
The results in Table 15 in the Appendix, where the impact of lifestyle is analysed separately for each age group, confirm the ones above. The decreasing pattern of F&V by age is clearly visible in the IV results where the coefficient decreases from 0.333 for the 16–24 years old to 0.254 for the 25–54 year olds and further to 0.203 for the 55–64 year olds. The decreasing impact of Sports Activity across age group is also clear.