Multiple regression allows for simultaneous statistical control of multiple covariates and is robust in dealing with various levels of measurement and moderate degrees of measurement error (Tabachnick & Fidell, 2001); it is thus an appropriate technique for this situation. The seven independent variables described above were regressed on each of the dependent variables (i.e. the factors) in turn, for a total of seven regressions. That is, in the first regression, the independent variables - current residence 1, current residence 2, church membership, time spent away from Tonga, gender, age and highest level of schooling – were regressed on the dependent variable value placed on common goals and projects over individual ones. The following six regressions include the same seven covariates each time, with only the dependent variable being replaced each time.

Each of the seven equations was found to be statistically significant. The results of the analysis are presented here first in narrative form and then in a summary table:

Value placed on common goals and projects over individual ones: Six explanatory variables influence responses to this scale. Those living overseas tend to score lower on this scale than those living in Tonga outside Nuku’alofa, controlling for the effect of all other variables in the equation (a condition that applies to the remaining results reported). Longer duration of overseas living also results in lower scores, as does membership in the Latter Day Saints church compared to those belonging to other churches. Older age is associated with higher scale scores. Of all variables in this model, level of schooling is most strongly associated with this scale, as Tongans with higher levels of schooling tend to score lower on this scale. (Level of schooling is either the most important or second most important explanatory variable for all seven scales, and is always negatively related, i.e. more education is related to lower scale scores.) The relative ranking of the remaining five explanatory variables, in descending order, is age, overseas current residence, time spent overseas, and church membership. (Gender has no effect on any of the seven dependent variables and is not included in the remaining results reported.)

Value placed on maintaining family relationships and cultural continuity: Responses to this scale are uninfluenced by current residence, or church membership. Age is related, again positively (older Tongans score higher than do younger Tongans in general), as is time spent overseas in the same manner as with the previous scale (negatively). Church membership is unrelated. In order of importance, in descending order, we have level of schooling, age and time spent overseas.

Appropriate faka’apa’apa in everyday, face-to-face relationships: Age, level of schooling, and time spent overseas are related to scores on this scale. As with the previous two dependent variables, level of schooling is negatively related and age is positively related. Current residence, and church membership are unrelated. Of the statistically significant covariates, time spent overseas is ranked as most important, followed by levels of schooling and age, respectively.

Obvious/performative aspects of Tongan identity: Level of schooling is negatively related with this scale, as is time spent overseas. Age is positively related. Current residence is unrelated, as is church membership. Level of schooling is in relative terms most influential, followed in order by age and time spent overseas.

Comparing quality and character of life in Tonga and overseas: While level of schooling and time spent overseas remain significant explanatory variables and in the same direction as in all regression equations discussed above, age drops out and current residence emerges as the dominant covariate. That is, Tongans living overseas are more likely than Tongans resident in Tonga but outside Nuku’alofa to score low on this scale. Likewise, but not to the same extent, those resident in Nuku’alofa tend to score lower than Tongans resident in Tonga but outside Nuku’alofa. Relative rankings of the explanatory variables are current residence overseas, level of schooling, time spent overseas and current residence in Nuku’alofa. There were significant differences between the ideas of people living overseas and people living in Tonga at the time they filled out the forms.

Attitudes about the hou’eiki as a people: Two variables are related to this scale: level of schooling (negatively) and church membership, also negatively. The order of importance is schooling followed by church membership. The effect sizes are relatively small, which is reflected in the low R-squared (.03) and F statistic and its associated significance level (f= 2.62; p < .05).

Iconic aspects of Tongan identity: Responses to this scale are influenced in this model by two variables: level of schooling (negatively) and current residence. Tongans resident overseas tend to score lower than Tongans in Tonga (outside Nuku’alofa). For the second time, current residence emerges as the most important influence in the model on the dependent variable.

Additional observations: Overall, current residence does not exert significant effects on four of the seven scales, i.e., it appears that place of residence—whether overseas, in Nuku’alofa, or elsewhere in Tonga—does not influence the value that Tongans place on maintaining family relationships and cultural continuity, their view of the appropriateness of faka’apa’apa in everyday face-to-face relationships, their views of the obvious aspects of Tongan identity, or attitudes about the hou’eiki as a people. Current residence is weakly related to value placed on common goals and projects over individual ones. The two exceptions are for the dependent variables comparing quality and character of life in Tonga and overseas, and iconic aspects of Tongan identity. Of the remaining variables, level of schooling is most important overall. Age exerts a significant influence on four dependent variables and time spent overseas is a significant influence on five dependent variables. Church membership has an effect in two equations, while (again) gender has no effect in any equation. These data are summarized, with appropriate statistical detail in Table 7-1, as shown.

Table 7-1: Multiple Regression Results for the Seven Models: Unstandardised Correlation Coefficients and Model Fit (R-squared)

* p </= .05; ** p </= .01; *** p</= .001

The data in Table 7–1 are presented as unstandardised correlation coefficients in order to help the reader to see the effect of variation in the independent variables (level of education, etc.) on the dependent variables. For example, we can see that for each difference in the level of education (1–4), there is a marked effect on factor 1, value placed on common goals and projects over individual ones, and with each increase in level of education (School) comes a corresponding decrease of -0.25 on a five-point scale; in other words, we can read the effect in the actual measure of the original data. It is important to recall here that this figure is for the effect of education within the multiple regression analysis, i.e., it is the effect of level of education when considered in the context of all the other independent variables at the same time.

The limitation of the presentation of the analysis using the unstandardised correlation coefficient is that we cannot assess the relative effect of the independent variables within a factor, because the scale of the effect varies. For example, we might think that there is a radically different effect from age than from level of schooling on factor one, but, because there are only four different values for level of education, and as many values for age as there are ages in the sample, the two are not comparable in relative terms. For this reason, the standardised correlation coefficient (called the beta) is usually calculated; the beta, however, measures relative effect across the independent variables (i.e. it is a standardised measure) but the values calculated are not directly comprehensible in terms of the original scale and are difficult to assess in terms of direct impact. In the interests of brevity, we present analysis of the beta in Table 7-2 showing the rank order of the independent variables on each factor.

It is this last table that shows most simply the effects of the independent variables. We have drawn out the detail in our narrative description, but a quick glance at the general pattern expressed in Table 7-2 indicates that ‘level of schooling’ and ‘age’ have the strongest and broadest relative effect in the analysis. We do not speculate or offer further opinion on the root causes of the patterns that have emerged here, but rather leave it to the careful reader to contemplate the analysis in terms of their own specific interests.

Table 7-2: Relative Importance of Statistically Significant Independent Variables within each Model, Based on the Standardised Correlation Coefficient