Spatial Effects Over Time-Framed Happiness

Abstract

The article investigates average marginal effects of intra-urban, urban-remote and perceived spatial differences over time-framed happiness. The study is based on a self-report social survey conducted in Adana which includes three different time frames for happiness, namely at present (short-term), in the last week and in the last 4 weeks (global). Over which, the effects of objective, subjective and social spatial variables through socio-economic and social capital variables are measured using logistic regression models. Based on cognitive neuroscience research findings, the expectations are that perceived aspects of life are more likely related to short-term happiness and objective aspects of life are more likely related to global happiness. The analyses reveal that urban-remote difference is more likely related to higher global happiness; vehicle dependent-others difference is more likely related to higher happiness; perceived spatial factors are more likely related to short-term happiness; lower relative income and higher neighborhood inequality are more likely to decrease global happiness; the unemployed and retired urban residents are more likely to feel less happy and related to global happiness; personal characteristics and socio-economic factors are more likely related to one-week happiness. The implications suggest that policies should be towards the city retirees and the unemployeds who feel less happy, and intensive public transportation areas and their residents who are the most unhappy and more disturbed by air pollution. For policy implementation, we recommend that the authorities discuss the public transport, distance to services, air pollution and unemployment issues, and adopt the retirement adjustment law to eliminate the grievances in the pensions of the retirees.

Appendices

Appendix 1

1.1 More on Theoretical Background

Neither does one day nor a short time make someone blessed and happy (Aristotle 1906). SWB perspective, have taken temporal differences into consideration by distinguishing (long-term) life satisfaction from (short-term) affect, but the relations between the short-term and long-term dimensions are often not adequately conceptualised (Nordbakke and Schwanen 2014). Dominance of present happiness in Western culture is claimed to underestimate both past and future (Şimşek 2009). The reference period of the last 4 weeks is argued to provide an adequate sample of feelings and experiences, rather than focusing on a short term that might be non-representative (Bradburn 1969; Diener et al. 2009; Ala-Mantila et al. 2018). On perception of time, Dolan (2014) reports that the perceived distance between now and 1 week from now is about the same as the perceived distance between one week from now and 4 weeks from now. In social surveys, respondents provide an evaluation of their past experiences, which is the judgements of a collection of various affects (Bremner 2011). The retrospective summary judgment of happiness Seligman and Royzman (2003) call it. The summary may include accomplishments and right choices which are important indicators of possible but uncertain future happiness. Hence, a time perspective should be included in social survey studies in order to capture retrospective judgements of respondents. These judgements are reliable and valid (Frey and Stutzer 2002; Kahneman and Krueger 2006; Diener et al. 2009) regardless of respondents being well educated or not. Diener et al. (2017) sheds some light to uncertainty of future happiness by indicating that circumstances and the choices people make in life influence their long-term SWB.

Researchers seem to avoid happiness assessment perhaps due to the implicit assumption that the key relationships are broadly similar across countries and regions (Stanca 2010) or due to fluctuations of happiness which make it difficult to differentiate the spatial difference (Wang and Wang 2016). Thus, they turn to Diener et al. (1985) satisfaction with life scale (Easterlin et al. 2011; Cramm et al. 2012; Aslam and Corrado 2012; Wang and Wang 2016; Ettema and Schekkerman 2016; Akaeda 2019), or to European social survey life satisfaction scale. On the contrary, in Western nations, studies at the cross-country and individual level show that happiness have considerable constancy (without much fluctuation) over time (Veenhoven 2006; Ehrhardt et al. 2000). It seems also that researchers combine happiness and life satisfaction assessments. Aşıcı and Eren and Asici (2017) used TURKSTAT life satisfaction scale “All things considered, how happy are you with your life?”. Ferrer-i-Carbonell (2005) used “How happy are you at present with your life as a whole?” for satisfaction with life in general assessment as subjective well being. Chyi and Mao (2011) and Asadullah et al. (2018) used “Generally speaking, how do you personally feel about your life?” on a scale of 1 to 5, where 1 = very unhappy, 3 = neither happy nor unhappy, and 5 = very happy. These three questions are trying to assess happiness based on general life. “all things considered”, “as a whole nowadays” and “generally about your life” seem to be non-instructive and unclear to respondents. Instead, depending on the focus of a study, specific time frames should be placed in the questions of happiness. For example, if “all things” means all important life experiences, then the assessment should be about life satisfaction. Happiness cannot capture “all things”, “whole nowadays”, “at present as a whole” or “generally about life” at the time of a social survey because happiness can not be measured as life satisfaction. On the contrary, life satisfaction can be measured based on happiness. Thus “how happy are you at present with your life as a whole?” is a vague question because it is asking how a person is happy with his/her life satisfaction. It expects respondents to remember momentary experiences in their life and make a subjective judgement, which is in contradiction with happiness definition of Kahneman (1999), and Seligman and Royzman (2003). However, this may have a particular importance to older people who remember only the most important turns in their lives such as weddings, born of children, accidents, deaths, buying a large house, finding good job. In that respect, Chyi and Mao (2011) seem to have the right question. Otherwise, “how happy are/were you [time frame]” (Ala-Mantila et al. 2018) and “how satisfied are you with your life” (Ehrhardt et al. 2000) are the right types of questions for the assessment of happiness and life satisfaction.

Appendix 2

2.1 Descriptive Statistics

Table 9 indicates that reported happiness decreases by time frame. Survey participants report higher mean happiness at present or short-term happiness.

Table 9 Descriptive statistics for SWB components

Pairwise Pearson correlations between measures of happiness are presented in Table 10. These correlations are in the range of 0.33–0.49, indicating a moderate to large overlap between the three measures. The lowest overlap is between 1WHAP and 4WHAP, and the highest overlap is between PHAP and 4WHAP.

Table 10 Pearson correlations among three measures of happiness

Table 11 reports correlations between time-framed happiness and independent variables. The significant correlations were low to moderate 0.0966-0.3826. PHAP shows the highest overlap with satisfaction with housing cost, 1WHAP with employment status and 4WHAP with health.

Table 11 Correlations between time-framed happiness and independent variables

Appendix 3

3.1 Sample Size, Reliability and Validity of Data

To find the sample size (n) required for the survey, the following formula is used:

$$n = \frac{{\chi^{2} Npq}}{{d^{2} (N - 1) + \chi^{2} pq}}$$

(1)

where χ² is table value (3.841 for 5% significance), N population size, p population ratio assumed to be 0.50 and d accuracy degree (margin of error) taken as 0.05. Substituting the values in (1) gives

$$n\, = \,\left( {3,841\left( {1747000} \right)\left( {0.5} \right)\left( {0.5} \right)} \right)/(\left( {0.05} \right)^{2} \left( {1747000 - 1} \right)\, + \,3,841\left( {0.5} \right)\left( {0.5} \right))\, \approx \,385.$$

Sample size should be minimum 385 with 5% margin of error. The minimum sample size requirement was met with 535 residents participating in the survey.

Cronbach alpha reliability test is conducted using SPSS to measure reliability of the questionnaire with 20 items (questions). Cronbach alpha reliability value is 0.903 which is acceptable. Removing a few corrected item-total correlation values (between 0.10 and 0.20) would not significantly increase the reliability value.

Construct validity was tested by principal component analysis applied to the questionnaire consisting of 20 questions. The rotation matrix was obtained by varimax method. KMO value 0.801 > 0.60 indicates that there are sufficient questions for each factor. The Barlett test significance level p = 0.000 < 0.001 shows that the correlation matrix is significantly different from the zero matrix. Total explained variance was 62%.

Appendix 4

4.1 Gini Coefficient

Coefficient values in Table 12 indicates that neighborhood inequality is the lowest in secondary pedestrian areas and the highest in intensive transit junctions.

Table 12 Summary statistics for neighborhood income inequality

Appendix 5

5.1 Logistic Regression Models

Binary logistic regression model (BLM) used for one-week happiness W_i in all models can be written as ln(p/(1-p)) where p is the probability of being happy. In these models, we assume that there exists an unobservable (latent) variable W^* where W represents the observed (respondent’s answer) dichotomous dependent variable. Cases with positive values of W^* are observed as W = 1, while cases with negative or zero values of W^* are observed as W = 0. The idea of a latent W^* is that an underlying propensity to well-being generates the observed state. While the propensity cannot directly be observed, at some point a change in W^* results in a change in what is observed (Long and Freese 2001). In logit models, the marginal effect for categorical variables indicates how P(W = 1) changes as the categorical variable changes from 0 to 1, keeping all other variables constant (at their values or averages). So, the marginal effect for a categorical variable x_k is P(W = 1|X, x_k = 1)–(P(W = 1|X, x_k = 0) (Williams 2012). Logistic regression assumes that the observations are a random sample from a population (i.e., i.i.d.) where the model is expressed as in (1), (2) or (3). Goodness of fit of a binary logistic regression model is tested by estat gof command (Archer and Lemeshow 2006) using Stata/IC 14.2. Marginal effects of x_k is calculated by ∂Pr(w = 1)/∂x_k = Pr(w = 1)*Pr(w = 0)*β_k (DeMaris et al. 1990:273) where x_k are OS_i, PS_i, SS_i, SE_i and SC_i, xβ = ∑_kβ_kx_k and Pr(w = 1) = 1/(1 + exp(-xβ)).

Ordinal regression model (logit) or ordered logit model (OLM) which essentially gives treshold estimates and a test of proportional odds assumption is run. α_i in (1) gives threshold values for i = 1 to j−1 where j is number of categories and i is the number of independent variables. Model specification is based on the test of parallel lines violation. Literature suggests to use either Generalized Ordered Logit Model (GOLM) or Multinom Logit Model (MLM) in case the assumptions of ordinal logit model are violated (Williams 2006). 4WHAP is explained by OLM which has a form logit[Pr(W ≤ j) = α_j–∑β_ix_i. The strength of the relationship is measured by McFadden’s pseudo R² statistics, which is based on the log-likelihood function for the model with the estimated parameters and the log-likelihood with the thresholds. Since pseudo R² values tend to be quite small compared to R² in multiple regression, values of 0.2 to 0.4 for rho-squared represent excellent fit (McFadden 1979). The higher pseudo R-squared indicates which model better predicts the outcome. Hence, increasing McFadden’s pseudo R² values as spatial variables added indicate improvement in model likelihood over the null model (Hemmert et al. 2018).

The use of a latent variable framework controls for measurement error in the dependent variable (Brereton et al. 2008) and for multicollinearity problems. Collinearity diagnostics is checked based on all correlations, significance of t-stat and F, variance inflation factor (vif), correlation matrix and, eigenvalues and condition index using SPSS, and robustness of standard errors is checked using Stata/IC 14.2.

Appendix 6

6.1 Demographic, Socio-Economic and Social Capital Variables over Time-Framed Happiness

Survey evidence suggests that attributes such as gender, age, marital status, income, employment, health and education are important indicators of SWB. Weakness of explanatory power of socio-demographic factors is well-known in the literature of SWB. In this study, socio-demographic factors explained about 12% of happiness.

The effects of human trust were more significant on higher 1WHAP. Human trust has a stronger effect than seeing relatives or friends more than 1–2 times a week on higher 1WHAP and on higher 4WHAP. The effect of seeing relatives or friends more than 1–2 times a week versus less than once a month was the weakest in explaining higher 1WHAP. Table 13 shows average marginal effects of socio-demographic variables over time-framed happiness. Results indicate that gender effect is not significant in explaining time-framed happiness. The effect of age category 54–65 relative to 18–29 shows spatial significance over 1WHAP. This effect is negative, stronger and more significant than PHAP and 4WHAP. The effect of being divorced or seperated relative to being married decreases the probability of 1WHAP by 23.9 pp when all spatial variables are included in the model. Household size shows no spatial significance. 

Table 13 Average marginal effects of socio-demographic variables over time-framed happiness

The relationship between locations and 4WHAP shows that health status was the most significant SE variable to explain 4WHAP. Hence, health status is included in all models of 4WHAP as a SE variable. SC is human trust, PS is the safety of neighborhood and SS is poverty or neighborhood inequality. The marginal effects of these explanatory variables over 4WHAP are calculated. The results indicate that a change from poor health to neutral status increases the average probability of being happy by 0.094 points or 9.4 pp; a change from poor health to good health status increases the probability of being happy by 30.7 pp. A change from “no human trust” to “human trust” increases the probability of being happy by 13.1 pp. A change from “being unsatisfied with neighborhood safety” to “being satisfied with neighborhood safety” increases the probability of happiness by 13.5 pp.

Table 14 shows that being unemployed versus currently employed and being retired versus currently employed are negative and significant in explaining time-frame happiness. When poverty line was replaced by Gini coefficient, the effects of employment status over time-framed happiness did not significantly change.

Table 14 Average marginal effects of socio-economic variables over time-framed happiness

Appendix 7

7.1 Objective Spatial Variables and Time-Framed Happiness

Residential area effects account for 5.4–10.2% of the variance in present happiness, for 8.9–15.3% of the variance in one-week happiness and 2.7–6.2% of the variance in four-week happiness.

City residence through health was more significant on 1WHAP than that through employment status, hence city residence effects were predicted through health. However, neighborhood safety and poverty effects were predicted through employment status due to higher significance. Overall, explanatory power was higher with employment status included in Models 1 of 1WHAP. Logistic model goodness of fit for 1WHAP with 535 observations and 315 covariates indicates that Pearson χ²(301) is 332.79 and χ² probability value is 0.1003. This test suggests that the model is a good fit.

Relative to reference category “remote neighborhoods”, the signs of AME where significant are as desired for PHAP and for 1WHAP. In Model 1, the marginal effects of OS over PHAP, as shown on the first column of Table 1 which are obtained from GOLM, indicates that, on average, those living in vehicle dependent neighborhoods are about 30% points (pp) more likely than those living in remote villages and towns to be happy at present. In other words, living in vehicle dependent neighborhoods increases the probability of being in higher PHAP by 0.300. The marginal effect of OS over 1WHAP, as shown on the first column of Table 2 obtained from BLM, shows that, on average, those moving from remote neighborhoods to intensive public transport areas increases the probability of being happy by 0.278. This increase slightly goes up when perceived spatial variable is added and goes down when social spatial variable is added. Moving from remote neighborhoods to any urban neighborhood increases the probability of being happy or very happy in the last week, i.e. Pr(W = 3).

Table 3 gives the outcomes for 4WHAP using OLM. The first column under 4WHAP indicate AMEs of the areas over higher four-week happiness through SE, the second column through SE, SC and PS, and the third column through SE, SC, PS and SS. The marginal effect of OS over 4WHAP, on the sixth column of Table 3, shows that, on average, those living in intensive transit junctions are 26 pp more likely than those living in remote neighborhoods to say they are happy in the last 4 weeks. Alternatively, spatial effects of moving from remote neighborhoods to intensive transit junctions increase the probability of higher happiness in the last 4 weeks by 0.26.

Health status reports lower spatial significance but higher explanatory power than employment status on 4WHAP. Through employment status higher 4WHAP shows the same spatial significance (< .001) in all regions except intensive public transport areas. However, spatial effect is around 5 pp stronger in each region except in secondary pedestrian areas where the effects are about the same. Through health 4WHAP shows higher spatial significance at intensive transit junctions (p < .01) and in secondary pedestrian areas (p < .05). In either way urban areas compared to remote are more related to 4WHAP on which the effects were predicted through health status for the sake of explanatory power.

Appendix 8

8.1 Environment Dissatisfaction, Perceived Spatial Variables and Time-Framed Happiness

All PS pairwise difference effects over time-framed happiness through control variables are reported in Tables 4, 5, 6 and 7.

Environment dissatisfaction effects account for 5.1–10.2% of the variance in present happiness, for 12.1–15.2% of the variance in one-week happiness and 2.5–6.2% of the variance in four-week happiness. Perceived spatial effects account for 7.0–11.9% of the variance in present happiness, 11.8–15.3% of the variance in one-week happiness and 4.1–8.5% of the variance in four-week happiness.

Table 4 shows the average marginal effects of environment dissatisfaction pairwise differences over higher PHAP. The pairwise effects of environment disturbances on higher PHAP relative to “no problem” are all negative as desired. Of which only distance to services—no problem (AME = − .274, − .199, − .191; p < .01, < .05, < .05) and poor public transport—no problem differences (AME = − .293, − .247, − .241; p < .01 all) are significant. On average, poor public transport versus no problem decreases the probability of being in happy or very happy category by 29.3 pp. Similarly, distance to services—air pollution (AME = − .209; p < .05) and poor public transport—air pollution differences (AME = − .228, − .194, − .197; p < .01, < .05, < .05) are significant. The effects of environment disturbances relative to “poor public transport” are all positive, and significant except for distance to services—poor public transport difference. Of which noise and traffic—poor public transport (AME = .235, .212, .208; p < .01 all), air pollution—poor public transport (AME = .228, .194, .197; p < .01, < .05, < .05) and no problem—poor public transport (AME = .293, .247, .241; p < .01 all) are significant. The effects of distance to services—noise and traffic difference (AME = − .217; p < .05) and poor public transport—noise and traffic difference (AME = − .235, − .212, − .208; p < .01 all) are negative and significant. The effects of environment disturbances relative to “distance to services” are all positive, and significant except for poor public transport—distance to services difference. The effects of noise and traffic-distance to services difference (AME = .217; p < .05), air pollution-distance to services difference (AME = .209; p < .05) and no problem-distance to services difference (AME = .274, .199, .191; p < .01, < .05, < .05). are positive and significant.

Table 5 shows the average marginal effects of environment dissatisfaction pairwise differences over higher 1WHAP. Significance of the pairwise effects of PHAP weakens or dissappears when compared to 1WHAP. Distance to services—no problem difference significance drops to p < .05 from p < .01 and to no significance from p < .05. AME increases 8.6 pp (from AME = − .274 to AME = − .184). The other pairs on this column remains insignificant. On air pollution column significance of distance to services—air pollution difference dissapears. However spatial significance of poor public transport—air pollution difference increases from p < .05 to p < .01 and but AME decreases about 2.6–6.0%. The other pairs on this column remains insignificant. On the poor public transport column, spatial significance of air pollution increases from p < .05 to p < .01, and positive spatial AME increases about 6 pp. Spatial significance of noise and traffic remains the same but positive spatial AME increases 7.1–7.4 pp. Spatial significance of no problem remains the same but positive spatial AME increases 3.1–3.4 pp. On the noise and traffic column, spatial significance of poor public transport remain the same, but negative spatial AME increases 7.1–7.3 pp. On the distance to services column, spatial significance of “no problem” dissappears.

Table 6 shows the average marginal effects of environment perception pairwise differences over higher 4WHAP. Significance of the pairwise effects of 1WHAP weakens or dissappears when compared to 4WHAP. Distance to services and air pollution columns show no spatial significance. On noise and traffic column, 1WHAP relative income significance of poor public transport-noise and traffic difference drops one level from p < .01 to p < .05, and negative spatial AME increases (from − 27.9 pp to − 13.7 pp) substantially 14.2 pp. This shows contribution of relative income to higher 4WHAP compared to 1WHAP. On poor public transport column, 1WHAP relative income significance of noise and traffic—poor public transport and no problem—poor public transport differences drops to p < .05. 1WHAP significance p < .01 of air pollution dissappears. On “no problem” column, 1WHAP relative income significance of poor public transport-no problem difference drops one level from p < .01 to p < .05, and negative spatial AME increases (from − 27.3 pp to − 14.6 pp) substantially 12.7 pp.

Table 7 shows the average marginal effects of perceived spatial pairwise differences over higher PHAP higher 1WHAP and higher 4WHAP. All higher happiness show positive spatial effects and significance of satisfaction with housing cost (p < .01) and satisfaction with neighborhood safety (p < .01). On average, being satisfied with housing cost versus dissatisfied increases the probability of being happy or very happy at present by 26.2 pp., and that of higher one-week happiness by 20.3 pp and that of higher four-week happiness by 16.1 pp. This shows that the probability decreases over time. The same can be said for satisfaction with neighborhood safety with a decrease at lower rates. These results indicate that although spatial significance remains the same, the average marginal effects show no contribution from objective and social spatial factors which lower the probability of higher happiness.

Appendix 9

9.1 Social Spatial Variables and Time-Framed Happiness

Social spatial effects account for 2.3–12.1% of the variance in present happiness, 7.2–15.3% of the variance in one-week happiness and 1.2–7.9% of the variance in four-week happiness.

Table 8 shows the effects of social spatial variables over time framed happiness. The results indicate that the effects of lower income are negative and significant on PHAP, 1WHAP and 4WHAP. When income changes from higher to lower category through employment status, residence location, social contact and safety of neighborhood, the probabilities of PHAP, 1WHAP and 4WHAP decreases by 9.8 pp, 11.9 pp and 9.9 pp, respectively; i.e. ∂Pr(PHAP = 3)/∂SS₁ = − 0.098 < 0, ∂Pr(1WHAP = 1)/∂SS₁ = − 0.119 < 0, and ∂Pr(4WHAP = 1)/∂SS₁ = − 0.099 < 0. On the other hand, the effect of Gini coefficient on 1WHAP is negative and significant on 1WHAP, suggesting that living in neighborhoods with more inequality is related to lower 1WHAP. More specifically, 1% increase in neighborhood inequality decreases the probability of higher 1WHAP by 23.9 pp. However, the effect of Gini coefficient on PHAP and 4WHAP shows no significance.

Appendix 10

10.1 Interaction Effects Over Time-Framed Happiness

The unemployed urban residents are more likely to feel less happy was tested by modelling time-framed happiness as a linear function of the interactions (between place of residence and employment status, and between place of residence and environment perception) for each component:

$${\text{W}}_{\text{i}}^{*} =\upalpha_{\text{i}} \, + \,\upbeta_{0} {\text{SE}}_{\text{i}} \, + \,\upbeta_{1} {\text{OS}}_{\text{i}} \, + \,\upbeta_{01} {\text{SE}}*{\text{OS}}_{\text{ij}} \, + \,\upbeta_{3} {\text{SC}}_{\text{i}} \, + \,\upbeta_{4} {\text{PS}}_{i} \, + \,\upbeta_{41} {\text{PS}}*{\text{OS}}_{\text{ij}} \, + \,\upbeta_{5} {\text{SS}}_{\text{i}} \, + \,\upvarepsilon_{\text{i}}$$

(2)

where SE*OS_ij is the i-th employment status in the j-th location and PS*OS_ij is the i-th perception in the j-th location.

Due to high unemployment rates in and around the city center, we expect the coefficient of interaction term to be negative, β₀₁ < 0.

The estimates of β₀₁ in Model (2) confirms our expectation that being unemployed in or around city center have adverse spatial effects on 1WHAP, i.e. β₀₁ < 0. Unexpectedly, the results also revealed that being retired in or around city center also had adverse spatial effects on 1WHAP. Being unemployed and living in central pedestrian (p < .005), secondary pedestrian (p < .05) and vehicle dependent (p < .05) areas were negative and significant. Being a retired resident in central pedestrian area was negative and significant (p < .01) in explaining 1WHAP. Being a retired resident in a intensive public transport area was negative and significant (p < .05) in explaining PHAP. There was no significant interaction effects on 4WHAP.

The estimates of β₄₁ in Model (2) confirms our expectation that air pollution in intensive public transport areas have adverse spatial effects on 1WHAP, i.e. β₄₁ < 0. Noise and traffic in secondary pedestrian areas was negative and significant (p < .01) in explaining PHAP. Distance to services in intensive public transport areas was negative and significant (p < .01) in explaining 4WHAP. Satisfaction with neighborhood safety in central pedestrian area and satisfaction with housing cost in heavy transit junctions both were positive and significant (p < .10) in explaining 4WHAP.

Using logistic models or OLS more interaction effects were investigated. Being in poor and moderate health status in central pedestrian areas, and being in moderate health status in transit areas have significant (p < .05 for all) adverse effects on PHAP. Only one age*location interaction effect was significant on PHAP. The coefficient was positive. One category increase in age (from 18–29 to 30–41 yrs) and a change of location from remote to transit area increases PHAP by 1.3 categories.

No interaction of any other demographic or socio-economic variable with residential area has significant effect over time-framed happiness. Human trust*65 + age interaction was negatif and significant (p < .05) on 4WHAP and human trust*54–65 age interaction was positive and significant (p < .05) on PHAP. Human trust*higher income interaction was negative and significant (p < .05) on 1WHAP. High income individuals who hesitate to trust people are less happy than low income people who do not trust people.

Appendix 11

11.1 Spatial Heterogeneity of Time-Framed Happiness

Spatial heterogeneity of time-framed happiness was tested by Model 1 based on significance and strength of spatial effects. We expect micro-level spatial heterogeneity, i.e., different time-framed happiness in different residential areas and different perceptions in different residential areas. The spatial heterogeneity of time-framed happiness was confirmed by the results that the higher short-term happiness is more likely in vehicle dependent neighborhoods; the higher one-week happiness is more likely in central pedestrian areas; and the higher four-week happiness is more likely in transit junctions. The estimates of interaction effects also confirms that air pollution in intensive public transport areas, noise and traffic in secondary pedestrian areas, distance to services in intensive public transport areas all have negative and significant effects over one-week, short-term and four-week happiness respectively. Satisfaction with neighborhood safety in central pedestrian area and satisfaction with housing cost in heavy transit junctions both were positive and significant in explaining four-week happiness. Satisfaction with housing cost was more related to short-term happiness and neighborhood safety was more related to four-week happiness.

Tables 1, 2 and 3 which also show different spatial effects on different time-framed happiness confirm that intra-urban time-framed happiness is spatially heterogeneous.