3.2.3. Analyses Based on Individual Data Comparison
To be more precise, we must examine the differences in the q
values of the individuals who participated in both dates of the two-day workshops. The number of data available was reduced to 80 in total: 42 in the C groups, and 38 in the F groups. Table 4
(a), below, shows the distribution of SVO types based on the first questionnaire, ‘b_SVO type’, as well as those based on the second questionnaire, ‘a_SVO type’. In order to account for the possible preference instability that may have occurred during the 10-day interval between the first and second date of two-day workshop, we classified the SVO types in the main analyses using ‘s_pro-social’ for those who are classified as pro-social twice, and ‘s_pro-self’ for those who were classified at least once as pro-self; the rest were classified by ‘s_other’. Table 4
(b) lists the number of participants in each of the three ‘s SVO’ types in the C and F groups.
The kind of individual data to be considered is the difference in q before and after the workshops, denoted by dq, where dq = [q after] − [q before] for each individual. Because both q after and q before must be valid for the same individuals, our sample size was further reduced to 63 for Q1 (30 for the C groups, and 33 for the F groups) and 65 for Q2 (31 for the C groups, and 34 for the F groups). Each workshop’s schedule and size were strictly constrained by the actual policy-making schedule within the municipality, and it was practically impossible to replicate the same workshops to increase our sample size.
depicts the average dq
of Q1 and Q2 in the C groups, as well as in the F groups. The dark blue bars show the average dq
of all of the participants; the light green bars are for the s_pro-social type; the yellow bars for s_pro-self type; and the gray bars for s_other type. There are four columns; the left two columns are for the dq
data from Q1 in the C groups (the first left column) and in the F groups (the second column). Similarly, the right two columns are for the dq
data from Q2 in the C groups (the third) and in the F groups (the fourth). Figure 8
shows a contrasting pattern between the C and F groups, which is mainly reflected in the s_pro-self type (yellow bars), where the average dq
of Q1 and Q2 in the C groups are positive, while in the F groups the average dq
of Q1 is zero, and the average dq
of Q2 is negative. This nominal pattern is consistent with the observation obtained in Figure 7
We conducted difference-in-differences ordinary least squares (OLS) regressions on individual’s dq
, as shown in Table 5
. Models (1) and (2) are for Q1, while (3) and (4) are for Q2. The regressors are a dummy variable for the F groups (F), a dummy for the pro-self type individual (s_pro-self), a dummy for the other type individual (s_other), along with two cross terms (s_pro-self * F) and (s_other * F), where the pro-social type is the baseline. The OLS regressions in all of the models (1) to (4) were carried out using robust standard errors corrected by clustering with 14 groups and gender.
Models (1) and (3) show simple results where the regressors do not include the interactions with ‘F’. Both models generated no significant results. This implies that the change in choice of q by the s_pro-social individuals was not different from zero in both the C and F groups. However, by adding the interactions with ‘F’, both models (2) and (4) detect a significantly positive effect of ‘s_pro-self’, coupled with a negative effect of its cross term with ‘F’. The effect of ‘s_pro-self’ is significant in both models, but the significance level of the effect of its cross term is less than 10% in model (2), and 15% in model (4). This means that the behavior of dq for the s_pro-self type individuals is different between the C and F groups. The change in dq of the s_pro-self type relative to the s_pro-social type within the C groups is positive and larger, which means that the s_pro-self type in the C groups became more myopic after the workshop, while the s_pro-social type did not experience a change in time perspectives. The negatively significant effect of the cross term of ‘s_pro-self’ with ‘F’ in model (2) (weakly in model (4)) cancels out the positive effect of ‘s_pro-self’. This means that the FD workshop prevented the s_pro-self participants in the F groups from becoming more myopic after the workshop relative to the s_pro-social participants.
reports the marginal effects of the probit regressions, where the dependent variable takes a value of one if dq >
0 for Q1 and for Q2, with the same regressors used in Table 5
. It is meaningful to examine the likelihood of dq
> 0, since an individual has to increase his/her switch point choice of q
by at least one unit of the question items if dq >
0. As in Table 5
, the regressions were carried out with the robust standard errors corrected by clustering according to groups and gender.
shows somewhat stronger results than Table 5
. This time, models (1) and (2) show a positive effect of ‘s_pro-self’, which implies that the s_pro-self individuals are more likely than the pro-social type to become more myopic after the workshop. By adding the interactions, model (2) detected a negatively significant effect of the cross term of ‘s_pro-self’ with ‘F’, which is not enough to offset the positive effect of ‘s_pro-self’, unlike model (2) in Table 5
.The FD workshop had an impact on the s_pro-self type individuals in the F groups, making them less likely to become more myopic compared to the s_pro-self type in the C groups, but the s_pro-self type individuals were still more likely to become more myopic after the workshop relative to the s_pro-social type. Table 6
reveals the worse adverse effect of workshop on the s_proself type than Table 5
. On the other hand, the effect of ‘s_pro-self’ crossed with ‘F’ is not significant in model (4). Thus, Table 5
and Table 6
provide observations that are consistent with the distinct behavior of the pro-self participants shown in Figure 7
and Figure 8
for Q1, while Table 5
weakly does so for Q2.
Next, we examined the dq
of individuals in subgroups defined by the participants’ characteristics of age and profession. Table 7
and Table 8
report the OLS regressions on dq1
, models (1) to (4) and dq2
, models (5) to (8) where the main focus is on the younger and older (=not younger) subgroups, and on the subgroup of city officers and the subgroup of the general public (=not city officers), respectively. We selected the first set of subgroups because we witnessed first-hand during the workshops that the older participants had less difficulty in acting as an imaginary future generation. This impression is shared by many of those who have conducted FD workshops. The second set corresponds to two sets of two-day workshops organized on different days. All of the models in Table 7
and Table 8
corrected standard errors by clustering with groups and gender.
The regression results shown in Table 7
have a dummy variable for the younger participants who are in their 20s and 30s (young). Models (1) and (5) have ‘F’, ‘young’, and their cross term as the regressors. The other models gradually add the other regressors used in Table 5
and Table 6
, and their cross terms with ‘young’. In all of the models, the constant term and ‘F’ are not significant. It follows that the participants in the older subgroup did not experience a change in their time perspectives before and after the workshop, regardless of the C or F groups.
Models (1) and (2) have no significant result. However, model (3) captures the positive effect of ‘s_pro-self’ and the negative effect of its cross term with ‘F’, which confirms the same result observed in Table 5
and Table 6
; that is, the s_pro-self participants in the C groups tend to become more myopic after the ordinary workshop relative to the s_pro-social participants, but their counterparts in the F groups did not exhibit such a tendency after the FD workshop. It is model (4) that detects a significant effect of ‘young’ in its cross terms. The significantly positive effect of ‘young’ crossed with ‘s_pro-self’ means that it is the younger subgroup among the s_pro-self participants whose time perspectives became shorter. Since our sample size is small, let us mention the negative impact of ‘young’ crossed with ‘s_prosef’ and ‘F’, which offsets the positive effect of ‘young’ crossed with ‘s_pro-self’, though its significance level is 13%. This weakly implies that the younger s_pro-self participants in the F groups did not suffer the adverse impact of the workshops that their counterparts in the C groups experienced.
On the other hand, the Q2 questionnaire captured the tendency among the younger subgroup in the F groups to obtain the longer time perspectives after the FD workshop, compared to the younger subgroup in the C groups relative to their older counterpart, as shown by the negative significant effect of the cross term of ‘young’ with ‘F’ in model (5). This effect remains negative (only weakly) in model (6). The indication of a change in time perspectives among the younger subgroup in the C groups appears in model (8) in its positively-significant effect of ‘young’ crossed with ‘s_pro-self’. Similar to the result in model (4), it was the s_pro-self participants within the younger subgroup that obtained shorter time perspectives after the workshop.
To investigate an effect of the difference in professions among the participants, we ran the OLS regressions on dq1
with the same regressors as in Table 7
but replacing ‘young’ with ‘officer’, a dummy variable for the city officers, as shown in Table 8
. In all four models (1) to (4) of the left half of Table 8
, the constant term is significantly negative, which implies that the base subgroup (=public) and particularly its pro-social type (the baseline in models (2) to (4)), obtained longer time perspectives after the workshop. Model (1) and (2) do not show any difference between the officer and public subgroups, but model (2) detects a positive significant effect of ‘s_pro-self’ indicating that the overall s_pro-self participants tend to become more myopic after the workshop relative to the s_pro-social participants. This effect remains positive in model (3) coupled with a significantly negative effect of its cross term with ‘F’, which again confirms our previous observations on the s_pro-self participants who became more myopic to the lesser extent in the F groups compared to the C groups in Table 5
and Table 6
. At the same time, model (3) captures a positively significant effect of ‘officer’ at a less than 11% level, which indicates that the city officers tend to become more myopic relative to the general public after the workshop. Model (4) shows that the positive effect of ‘officer’ in model (4) comes from the positive effect of ‘officer’ crossed with ‘s_pro-self’, which means that such a change in the time perspectives of the officer subgroup is brought about by the s_pro-self type individuals within the officer subgroup. This effect is further divided into the effect of whether an individual is part of the F group or not, as depicted by the negative effect of ‘officer’ crossed with ‘s_pro-self’ and ‘F’. In this case, the total effect of the cross term of ‘officer’ with ‘s_pro-self’ and its cross term with ‘F’ is negative and significant at a less than the 1% level, which implies that the s_pro-self participants among the city officers in the F groups obtained longer time perspectives after the FD workshop relative to the pro-social baseline type than their counterpart in the C groups.
In contrast, the constant term in models (5) to (8) is not significant, such that Q2 questionnaire did not elicit a change in the time perspective of the base subgroup and its baseline pro-social type. Model (7) shows the same tendency for ‘s_pro-self’ and its cross term with ‘F’, as in model (3), but the effect of ‘officer’ is positive only nominally in model (7). Model (8) generates the same result as in model (4), only with a slightly weaker significance level.
As seen in models (4) and (8), the tendency of the s_pro-self individuals to become more myopic after the workshop is more pronounced among the city officers relative to the general public. The s_pro-self individuals among the city officers in the F groups could not only avoid becoming more myopic but also obtained longer time perspectives. Such a tendency is not observed for the general public in either the C or F groups. This difference between the city officers and the public subgroups does not seem to have stemmed from the young vs. old factor examined in Table 7
, because the percentage of younger participants was about 50% in both subgroups.
Through the analyses up to this point, there seems to have been some difference in what Q1 and Q2 can detect. In Table 5
and Table 6
, the absolute size of the coefficient associated with ‘s_pro-self’ and its cross term with ‘F’ on Q1 is nominally larger than those on Q2. So too are the absolute size of coefficient of ‘s_pro-self’ crossed with ‘young’ and its cross term with ‘F’ in Table 7
. The cross-model Wald test detected that the coefficient of ‘s_pro-self’ crossed with ‘officer’ and its interaction with ‘F’ in Table 8
is larger in the Q1 regression (model (4)) than the Q2 regression (model (8)), with significance levels of less than 7% and 0.3%, respectively. This is indicative that the pro-self individuals were more sensitive to the individual choices in Q1 than the non-individual choices in Q2.
Furthermore, there are things that only Q2 revealed. A positive impact of FD on younger participants in model (5) of Table 7
was detected only by Q2. The behavior of the s_other type participants depicted in Figure 8
is captured only by model (4) on Q2 (as reflected in the negative coefficient of ‘s_other’ crossed with ‘F’ in Table 5
and Table 6
), and not by Q1 of both tables. This means that those who are neither the s_ pro-social nor s_pro-self type obtained longer time perspective after the workshop compared to their counterparts in the C groups relative to the s_pro-social participants. Since the number of s_other type individuals is very small, we do not discuss the s_other type participants much further, but we will come back to the difference of Q1 and Q2 in the next section.