2. Data Collection and Descriptive Results
Between 1990 and 2006 Bellingham sent 32 staffers and Tateyama 34, all were invited to participate in the study in early 2007 [
32]. From this universe Bellingham provided 19 respondents a participation rate of 59% and Tateyama 27 or 79% participation. As noted earlier the full study had two purposes, first to determine if a common structural framework exists for the evaluation of current QOL what we refer to as necessary conditions which are reported on here, and second to determine if there was an infusion of cognitive social capital, hence the successful demonstration of meeting sufficient conditions, to be reported on in a latter paper. For the entire study a four round Delphi was conducted, of which only the first round data is utilized here. During the study all interviews were carried out following the rules of the Declaration of Helsinki of 1975 (
http://www.wma.net/en/30publications/10policies/b3/), revised in 2008. Approval for the study instruments and was obtained from the Western Washington University Human Subjects Review Committee in June of 2006. Written informed consent was obtained from each participant prior to their inclusion in the study.
In the first round of the multi-round Delphi study [
33], panelists were asked to evaluate and discuss 11 specific indicators of QOL for each city on a scale of 1 to a high of 1,000 as well as provide an over-all value of QOL in each city, item 12 (see
Table 1 for a complete list of QOL indicators). The indicators were created based on Costanza
et al.’s definition of QOL and grouped into three broad categories reflecting the physical, service, and social environment [
34]. Average results by each panel for each city are provided in
Table 1, along with measures of significant differences from the Wilcoxon signed rank sum test, and a summary symbol indicating which city scored higher when significant differences are found.
From the center columns of
Table 1 it is clear that the Bellingham panel rates all the indicators and the summary measure of QOL for itself equal to or significantly [
35] higher than Tateyama except for the measure of public safety, a not surprising result. The significantly higher values for housing, parks, leisure, education, and shopping may be attributed to basic differences in things like population density, Bellingham’s prime location for recreation between mountain and sea, the existence of a comprehensive university in Bellingham, and the city’s role as a major retail outlet for Canadian cross-border shoppers from Canada’s third largest metropolitan area. However, the interesting result is that the Bellingham panel ranks transportation equal in both places despite a considerable difference in infrastructure and modal choices. Likewise the Bellingham panel finds the social environment equal in both places.
The Tateyama panel results (right side of
Table 1), collected completely independently, generally agree with the significantly higher values noted above with one exception, shopping, which they rank as equal between the two cities. However, by looking at the mean values it is clear that this is not a sign that the Tateyama panel thinks that Tateyama has better shopping opportunities then identified by the Bellingham panel, but rather that they may be less impressed by or knowledgeable of the opportunities in Bellingham. The most interesting thing from the Tateyama panel is they rank the quality of their transportation and social environment significantly lower than Bellingham’s.
Table 1.
Average of QOL indicators by panel between cities and Wilcoxon Signed Ranked (WSR) Test results.
Table 1.
Average of QOL indicators by panel between cities and Wilcoxon Signed Ranked (WSR) Test results.
| Bellingham Panel | Tateyama Panel |
---|
Bellingham | Tateyama | WSR results | Summarya | Bellingham | Tateyama | WSR results | Summarya |
---|
Group | Indicator | Mean | Mean | sig. | | Mean | Mean | sig. | |
Physical | 1. Quality of Housing | 821 | 711 | 0.028 | B** | 794 | 576 | 0.000 | B*** |
Physical | 2. Quality of Transportation | 708 | 781 | 0.188 | --- | 719 | 520 | 0.000 | B*** |
Physical | 3. Parks and Outdoor Recreation Opportunities | 909 | 687 | 0.001 | B*** | 822 | 470 | 0.000 | B*** |
Services | 4. Opportunities for Leisure and Amusement | 795 | 736 | 0.083 | B* | 800 | 502 | 0.000 | B*** |
Services | 5. Opportunities for Educational Enrichment | 822 | 738 | 0.075 | B* | 736 | 563 | 0.002 | B*** |
Services | 6. Opportunities for Shopping | 800 | 624 | 0.003 | B*** | 679 | 643 | 0.191 | -- |
Services | 7. Feeling of Personal Safety and Security | 721 | 861 | 0.007 | T*** | 642 | 761 | 0.001 | T*** |
Social | 8. Environment for Families | 808 | 787 | 0.330 | --- | 743 | 670 | 0.025 | B** |
Social | 9. Environment for Children | 776 | 784 | 0.725 | --- | 728 | 674 | 0.016 | B** |
Social | 10. Environment for Retired People | 761 | 716 | 0.609 | --- | 723 | 663 | 0.053 | B* |
Social | 11. Feeling of Community and Volunteer Spirit | 746 | 707 | 0.929 | --- | 775 | 604 | 0.001 | B*** |
Over-all | Over-all Quality of Life | 853 | 758 | 0.023 | B* | 767 | 650 | 0.001 | B*** |
Looking at these same results in a slightly different fashion by directly comparing the two panels’ evaluations for each city (
Table 2) and using the Wilcoxon-Mann-Whitney test produces a rather strong agreement concerning Bellingham but quite a bit of difference regarding Tateyama. In evaluating Bellingham only three indicators plus the summary QOL measure are significantly different with the Bellingham panel assigning each of these higher numbers; these are Parks, Shopping, and Safety. Given the fact that Bellingham ranks near the top in park acreage per capita for Washington State, it is not surprising that the Bellingham panel would provide a higher value than Tateyama visitors. Likewise we have already discussed why shopping values might be different. Finally, in regards to safety the higher value from the Bellingham panel might indicate a different relative perspective on what is acceptable from an American viewpoint, thus celebrating the relative safety of Bellingham.
Table 2.
Average of QOL indicators for each city by different panels and Wilcoxon-Mann-Whitney (W-M-W) Test results.
Table 2.
Average of QOL indicators for each city by different panels and Wilcoxon-Mann-Whitney (W-M-W) Test results.
| Bellingham Values | Tateyama Values |
---|
Response from | | | Response from | | |
---|
Bellingham panel | Tateyama panel | W-M-W results | Summarya | Bellingham panel | Tateyama panel | W-M-W results | Summarya |
---|
Group | Indicator | Mean | Mean | sig. | | Mean | Mean | sig. | |
Physical | 1. Quality of Housing | 821 | 794 | 0.562 | --- | 711 | 576 | 0.036 | Bm** |
Physical | 2. Quality of Transportation | 708 | 719 | 0.685 | --- | 781 | 520 | 0 | Bm*** |
Physical | 3. Parks and Outdoor Recreation Opportunities | 909 | 822 | 0.085 | Bm* | 687 | 470 | 0.001 | Bm*** |
Services | 4. Opportunities for Leisure and Amusement | 795 | 800 | 0.718 | --- | 736 | 502 | 0 | Bm*** |
Services | 5. Opportunities for Educational Enrichment | 822 | 736 | 0.131 | --- | 738 | 563 | 0.003 | Bm*** |
Services | 6. Opportunities for Shopping | 800 | 679 | 0.009 | Bm*** | 624 | 643 | 0.645 | -- |
Services | 7. Feeling of Personal Safety and Security | 721 | 642 | 0.054 | Bm* | 861 | 761 | 0.028 | Bm** |
Social | 8. Environment for Families | 808 | 743 | 0.148 | --- | 787 | 670 | 0.009 | Bm*** |
Social | 9. Environment for Children | 776 | 728 | 0.434 | --- | 784 | 674 | 0.035 | Bm** |
Social | 10. Environment for Retired People | 761 | 723 | 0.657 | --- | 716 | 663 | 0.266 | -- |
Social | 11. Feeling of Community and Volunteer Spirit | 746 | 775 | 0.551 | --- | 707 | 604 | 0.067 | Bm* |
Over-all | Over-all Quality of Life | 853 | 767 | 0.022 | Bm** | 758 | 650 | 0.005 | Bm*** |
When turning to Tateyama, the Tateyama panel consistently ranks itself lower than the Bellingham panel on all but two measures, shopping (discussed previously) and environment for retirees. The equality of the retirees’ environment measure seems to be a result of the Bellingham panel ranking this measure quite low, considerably lower than the environment for family or children rather than the Tateyama panel scoring it high. Clearly the Tateyama panel is much more critical of or perhaps modest about itself then the visitors which is also affirmed in the summary over-all QOL measure.
Based on these descriptive results from each city a number of general conclusions can be drawn here: (1) Bellingham emerges from this set of indicators as generally equal to or superior to Tateyama except for public safety as scored by both panels and (2) the Bellingham panel tends to assign higher scores although not always significantly higher, but this is most notable when the two panels evaluated Tateyama. This last point might be an indicator of cultural bias which has been seen in other studies using Likert like scales [
36]. Although this descriptive analysis provides insight into individual similarities and differences, it only does so in a bivariate fashion. Our next task is to apply a multivariate technique to this data to discover if it represents random and unconnected variations or a systematic yet different positions on the same QOL framework.
3. Methodology: Proportional Odds Modeling with a Cumulative Logit Link
To search for a common latent structure, a type of ordinal regression rather than linear regression was chosen. Proportional odds modeling with a cumulative logit link was used to model the ratings of our pair of sister cities on twelve QOL indicators. Forms of proportional odds modeling has been widely used in QOL studies especially in the medical field [
37,
38], but less so in looking at urban QOL issues [
39,
40]. However, the focus of such earlier work has been different; it has been based on creating explanatory models where a series of independent variables are used to explain ordinal dependent QOL measures. Here the focus is not to explain but to compare the results from two groups of respondents to look for the existence of a common latent QOL structure identifiable to and used by both groups. As noted above, there were twelve QOL questions, each of which was asked twice: once about Tateyama and once about Bellingham. Individuals were asked to respond to each question with a value between 0 and 1000. The great majority of individuals gave responses of 100, 200, 300, . . . , 1000, as if this were an ordinal response variable with 10 or fewer categories. As a result, we decided that it would be most appropriate to treat the data as ordinal responses and choose ordinal regression over the more common linear regression [
41]. Second, given the scarcity of data below a value of 500, and very few values that did not fall exactly on a hundred entry, we decided to reduce the number of categories. We binned the responses as follows: a raw score between 0 and 500 was coded as rank 1, the lowest; a score in the 501–600 range was coded as a 2, a score between 601 and 700 was coded as a 3, and so forth, with a score between 901 and 1000 coded as a 6. This gave a total of 242 1’s, 115 2’s, 164 3’s, 270 4’s, 187 5’s, 90 6’s, and 36 missing responses. We felt this binning scheme that collapsed values of 0 to 500 into a single bin and all others into their own based on steps of one hundred accurately reflected the over-all response pattern [
42]. See Equation (1): Two city QOL proportional odds model with a cumulative logit link:
In applying our model (Equation (1)) to test if individuals from Tateyama and Bellingham gave the same or systematically different responses to the QOL questions the following variables were included in the model: (1) the respondent’s city (Xi, which is 0 if individual i is from Tateyama and 1 otherwise), (2) the city being asked about (Wm, which is 0 if city m is Tateyama and 1 otherwise), and (3) which of the 12 questions was being asked (to limit clutter and as is a common way of presenting such a model, in equation 1 we provide no specific symbol for this variable, instead the equation itself is written for a specific k or question, meaning that a vector of twelve of these equations are utilized), and (4) the dependent QOL indicator’s ordinal score (Yikm, which refers to the rank score given by individual i for question k about city m). (5) We also included an interaction term between the question being asked and the city the question was asked about (Xi Wm, which equals 0 except for a case of individual i coming from Bellingham and evaluating city m, Bellingham, where it equals 1). Finally, (6) to account for correlation between responses from a single individual, we included a random effect for each individual. This random effect, ui, was assumed to follow a normal distribution with a mean of 0 and a variance σ
2. In this way, we were able to look at the twelve quality of life questions in aggregate and draw conclusions about overall patterns [
43].
Thus, in our model, the log-odds that individual i will give a response less than or equal to rank j when asked question k (k = 1, 2, . . . , K) about city m is given by equation 1 above.
Solution of this proportional odds model requires relative calibration, in this case around one of the K questions. We selected Question 12, an estimate of overall QOL, as our baseline around which all other QOL measures are arranged. This results in β12 and γ12 being constrained to be zero. In addition, since it simultaneously fits logistic regressions for all choices of j (j = 1, 2, . . . , J − 1), no regression is fit for j = J (or j = 12 in our particular case) because P(Yikm ≤ J) = 1.
The model estimates six parameters. The first θj acts as a set of monotonically increasing intercepts or thresholds. As the sum of the ordinally ranked evaluations j increase from 1 through J-1, so too does θj. Next, βk captures any individual differences between QOL measure k (Question k) and the baseline measure, in our case the overall QOL measure from Question 12. Thus, the individual measure k can be equal to Q12 or significantly higher or lower than this baseline. On the other hand, γk captures individual differences in QOL measure k between Tateyama and Bellingham. While both cities are assumed to occupy positions on the same QOL framework, differences in resource endowments and policies will cause variations in position on the structure for different indicators. Then α1 captures a systematic shift in the ordinal response by panelists from one city versus the other, what we have assumed to be a cultural shift as discussed in the previous section. As noted, it has been well documented in the literature that culture can cause such a systematic shift in ordinal responses when evaluating the same entity. Next α2 captures a systematic across the board resource or policy shift (as opposed to a specific one captured by γk above) in QOL between two target cities. Basically this indicates that one city has a generally higher/lower QOL than the other, thus indicative of each city having different positions on the latent structure based on a systematically different set of resource endowments and/or policies. Finally, δ captures a relative interaction shift that results from a respondent evaluating their own city versus the other, potentially reporting a systematically higher or lower value than would otherwise be expected. It is conceptualized as capturing additional knowledge resulting from greater intimacy with one’s own area that can only be accomplished by an individual who has been integrally involved over the long term with a specific place, knowledge which is not duplicable during a short visit. Thus, it can only be applied by respondents to their own place of residence. Also, as a relative term it can be estimated by focusing either on Tateyama or Bellingham. Either way δ would produce the same numeric value but in each case respectively an opposite sign would appear.
Analysis was run using the Christensen’s ordinal package in R [
44].
5. Conclusions
In the era of globalization urban networking is a growing trend turning former competitors into cooperators. Yet, as noted earlier, it is one thing for cities to network, but a completely different issue to derive value from such connections especially through social capital development and utilization by means of a UNKN. In nurturing such learning cities in pursuit of sustainability and high QOL, both necessary and sufficient conditions must be met in order for knowledge and value transfer and utilization to occur, particularly through cognitive social capital development. Using Proportional Odds Modeling with a Cumulative Logit Link to explore underlying latent structure for a two city SCTR professional exchange, this study has demonstrated that necessary conditions have been met—that there is a common QOL framework upon which each city falls despite their great cultural, resource, and policy differences. What remains for future work is demonstration that sufficient conditions are also met, that actual knowledge transfer occurs, and that value is realized from networking. Again, in prior studies this too was assumed to occur if indirect measures of structural social capital development were demonstrated. Thus, this study addresses one lacuna, the meeting of necessary conditions for cognitive social capital development from SCTRs, but leaves demonstration of meeting sufficient conditions as the next step. Other areas that also warrant additional exploration would be exploring relationships between cities of different sizes, cultures, and levels of economic development.