Interim Survey Data Analysis: Hypotheses and Analysis of Models

Sections: Introduction -> Survey Details -> Sample Information -> Survey Items -> Hypotheses and Analyses of Models -> Conclusion

 

As the overarching title of this post suggests, the analysis presented here should be treated as interim. Many of the variables outlined above are approximations of concepts and require considerable refinement, particularly in terms of subjecting their components to confirmatory factor analyses. The use of multiple regression is also a starting point, with the aim being to move onto structural equation modelling in the hope that it can accommodate the complexity of the theoretical models. Still, the results provide some food for thought in the sense that they identify variables in all four elements of the model that are robustly important in accounting for levels of political activity.

The main purpose of this analysis, before identifying consistently significant independent variables, is to consider how well the four elements of the full model account for political activity. This is achieved, first, by cumulatively adding those elements to the model and observing their impact on the Adjusted R-Square (more conservative than its non-adjusted counterpart), or proportion of the variation in the dependent variable accounted for by the independent variables. As such the hypotheses that are being tested in the interim analyses are as follows:

 

Hypothesis 1 – Adding the variables representing the forms of capital will improve the Civic Voluntarism Model in terms of the amount of variation in the dependent variable accounted for;

 

Hypothesis 2 – Adding the variables representing perception of privilege will improve the Civic Voluntarism Model in terms of the amount of variation in the dependent variable accounted for;

 

Hypothesis 3 – The Bourdieusian capital and perception of privilege model of political participation will perform as well as the Civic Voluntarism Model in terms of the amount of variation in the dependent variable accounted for.

 

The results of the cumulative analyses are shown in Table 9, which presents the variables relating to background characteristics as the baseline model. The components of the Civic Voluntarism Model were then added in the second model, followed by the additional Bourdieusian economic, cultural, and social capital variables in the third model. Finally, the variables relating to perception of privilege were added to create the full model.

Despite including variables that are statistically significant (highlighted in bold and italics) at the 10% level or above in terms of their impact on political activity, the baseline model (Model 1) does not perform well in terms of Adjusted R Square (0.083) and accounts for less than 10% of the variation in the dependent variable. Things get more interesting when the variables from the Civic Voluntarism Model are added (Model 2). Again, some of the variables are significant at the 10% level or above but the real action is in the impact on Adjusted R Square, which shoots up to 0.460. This means that, statistically speaking, the components of the Civic Voluntarism Model account for a full 37% more of the variation in political activity than do background characteristics alone.

Less impressive is the effect of adding the Bourdieusian elements of economic, cultural, and social capital to the model (Model 3). Some of those variables are significant at the 10% level but, combined, they only increase the Adjusted R Square to 0.486. The 13 variables in the Bourdieusian element of the model, then, only account for 2.6% more of the variation in the dependent variable than do background characteristics and the Civic Voluntarism Model alone. Finally, the addition of the three variables relating to perception of privilege (Model 4) raises the Adjusted R Square only a tiny amount (to 0.489). These results are summarised in Chart 3, in which each iteration of the model includes an indication of the number of variables in it (in parentheses), and the full model is labelled as ‘Civic Voluntarism, Bourdieusian, and Perception’.

 

Table 9 – Cumulative Addition of Elements to Construct

Full Model Accounting for Political Activity

(please click on the table for a full-size version)

2015-08-13 Corrected Interim Survey Data Analysis Table 09

 Chart 3 – Adjusted R Squares of the Cumulative Models

2014-10-14 Interim Survey Data Analysis Chart 03

 

The fifth model in Table 9, labelled as ‘Parsimonious Full’ in Chart 3, includes only the significant (at the 10% level) variables from the full model. As can be seen this model includes fewer variables than all of the others except the baseline model but it performs just as well as the full model in terms of Adjusted R Square (0.493). This means that with just 13 variables we can account for over 49% of the variation in the dependent variable. Still, the parsimonious model performs only slightly better than the Civic Voluntarism Model, which seems to account for most of the impressively high Adjusted R Square. However, from the cumulative models we cannot observe how well the elements of the full model perform on their own and, therefore, in comparison to each other.

Of particular interest when treating the elements of the full model as separate models in their own right is the performance of the Civic Voluntarism Model compared to that of the Bourdieusian and perceptual model. This is because the former can be considered the established model whereas the latter is a new one drawing on the theories outlined in the literature review. Each model is tested in various forms but the variables from the baseline model are included in all instances. This is because they are widely used control variables and because they represent important background characteristics in their own right. The results of these comparisons appear in Table 10.

The focus again is on amount of variation in the dependent variable, which remains the frequency of political activity, that is accounted for by the independent variables in each model. The first model (Model 2a) is the Civic Voluntarism Model with the group recruitment variable removed. This has been done because the power of that variable (Beta = 0.445) may indicate problems with it. In particular, the question that it is based on is only separated from the question that underpins the dependent variable by three other questions. Further, those three questions are themselves about levels of political or voluntary activity, and requests to engage in such behaviour. As such there is a risk that respondents, having just answered questions on their levels of political and voluntary activity, will have subconsciously altered their answers to proximate questions on recruitment so that they reflected those levels of activity more closely. The power of the group recruitment variable can be seen in the reduction of Adjusted R Square (from 0.460 to 0.301) when it is removed from the Civic Voluntarism Model (compare Model 2 with Model 2a), and it is for this reason that the subsequent models are tested both with and without it included.

The parsimonious Civic Voluntarism Model (Model 2b), including the significant variables from the full equivalent, is tested next. Given that it only includes eight variables it has a remarkably high Adjusted R Square (0.463), though a large proportion of that comes from the inclusion of the group recruitment variable. Still, this is by far the best performing model if the number of variable included is considered alongside the Adjusted R Square. By contrast, the model including the variables representing Bourdieu’s economic, cultural, and social capital as well as the perception of privilege variables (Model 4b) accounts for 20% less of the variation in the dependent variable (Adjusted R Square = 0.264) than does the parsimonious Civic Voluntarism Model. Even if the recruitment variable is included alongside the Bourdieusian and perception variables (Model 4c) they do not perform as well as the Civic Voluntarism Model (Adjusted R Square = 0.434). The parsimonious versions of the Bourdieusian and perceptual models excluding recruitment (Model 4d) and including it (Model 4e) perform only marginally better (Adjusted R Squares of 0.273 and 0. 438 respectively).

 

 Table 10 – Comparison of Elements of the Full Model

as Separate Models Accounting for Political Activity

(please click on the table for a full-size version)

2014-10-14 Interim Survey Data Analysis Table 10

 

The results of the model comparison are summarised in Chart 4, which again indicates the number of independent variables in each model (in parentheses), and includes the full Civic Voluntarism Model (from Table 9) for reference. The Bourdieusian and perception models perform consistently worse than the various iterations of the Civic Voluntarism Model. If a conservative approach is taken and the group volunteering variable is excluded (because of the question’s proximity, in the survey, to the question underpinning the dependent variable) then the gap is narrowed to a very great degree. Indeed, it accounts for between 2.8% and 3.5% less variation (respectively, in its parsimonious (Model 4d) and non-parsimonious (Model 4b) forms) in the dependent variable than does the Civic Voluntarism Model excluding the group recruitment variable. This gap is sustained if the variable is included in the parsimonious versions of both (Model 2b and Model 4e). However, if a strict reading of the texts that inform models is adhered to then the group recruitment variable should only be included in the Civic Voluntarism Model, making it much stronger than its Bourdieusian and perceptual counterpart.

 

 Chart 4 – Adjusted R Squares of the Comparative Models

2014-10-14 Interim Survey Data Analysis Chart 04

 

In terms of the three hypotheses, the first two are supported to a very small degree whilst there is no evidence emerging to support the third. That is to say that adding the Bourdieusian and perception of privilege variables do improve the Civic Voluntarism Model, marginally, in terms of the amount of variation in the dependent variable that is accounted for. However, the number of variables added to gain such a small increase in Adjusted R Square suggests that the Civic Voluntarism Model alone is preferable. This is further supported by the fact that none of the analyses indicate that the Bourdieusian and perceptual model alone performs better than the Civic Voluntarism Model.

Although the preceding seems pretty conclusive there are hints of other paths that may be worth following. The fact that the Bourdieusian and perception of privilege models perform only slightly worse than the Civic Voluntarism Model suggests that they may be variables of interest included in the former, which is confirmed if Table 9 is reviewed in more detail. This reveals that four of the variables in the baseline (background characteristics) model remain significant at the 10% level or above regardless of the other variables that are added to the model. Taking the results of the parsimonious full model are it appears, first, that respondents who did not identify as heterosexual were more likely to be politically active than those who identified as heterosexual (Beta for the heterosexual binary variable was -0.067). Although the Bs are difficult to interpret because the dependent variable is the sum of eleven different political activities, it can still be observed that heterosexual respondents had an average score that was over two points lower than their counterparts who did not identify as heterosexual (B = -2.276).

Still focusing on the parsimonious full model, having some level of limitation due to a disability was also related to a greater level of political participation (Betas for the binary variables indicating a little limitation and a lot of limitation were, respectively, 0.051 and 0.065). Respondents with a little limitation, on average, had a dependent variable score that was more than one point higher than respondents with no limitation stemming from a disability (B = 1.130) whilst those respondents with a lot of limitation had a score that was a approaching two points higher (B = 1.849). Similarly, having a larger number of health conditions was positively related to political activity (Beta = -0.065). The negative sign preceding the numbers relating to this variable appear because it has been inverted so that a higher score equates to being healthier (i.e. having less health conditions). Thus, amongst the respondents to the survey, having an additional health condition was related to an increase of almost three quarters of a point on the dependent variable scale (B = -0.665, again the negative sign is present because the independent variable is inverted). These findings may appear counterintuitive and any interpretation at this stage is speculative, however all of the groups that have higher levels of political activity either have a history of being socially marginalised, or have practical issues in their day-to-day lives that may prompt them to be more active in pursuing influence.

Moving on to the active components of the Civic Voluntarism Model we can see that work based civic skills are positively related to political activity, though the effect is small (Beta = 0.062) and a one point rise on the civic skill scales is associated with a rise of less than a tenth of a point (B = 0.080) on the dependent variable. By contrast, political interest has a much stronger positive relationship with political activity (Beta = 0.272) and, in fact, is the second strongest predictor in the parsimonious full model. On average a one point rise on the combined political interest scale is associated with approaching a half point (B = 0.457) rise in the rough measure of political activity used here. Only group recruitment has a stronger positive relationship with political activity (Beta = 0.411) than political interest. In fact, as mentioned, group recruitment is by far the strongest predictor of political activity, and a one point rise on the group recruitment scale is associated with over a half a point rise in the dependent variable (B = 0.547). The three active components of the Civic Voluntarism Model, then, all have the anticipated positive relationships with the dependent variable, and political interest and group recruitment are the two variables in the parsimonious full model with the strongest relationships with political activity.

As can be discerned from the comparisons of the various models above, the active components of the Bourdieusian model are less powerful than those in the Civic Voluntarism Model. Nevertheless, there are some interesting relationships at play. The count of cultural tastes variable has a negative relationship with political activity (Beta = -0.116), which is to say that having more encompassing cultural tastes is associated with being less politically active. This is not an especially strong relationship as it is a non-integer variable running from zero to three, so a full one point change on the scale is a great deal of movement. Still, such a change is associated with something approaching a one and a half point lower score on the dependent variable (B = -1.399). Slightly stronger is the impact of the frequency of cultural activities (Beta = 0.161), which is positively related to political activity. This is a summary variable with a long scale and a one point change on it is associated with only a very small change on the dependent variable (B = 0.077). Together, the impact of these measures of cultural capital suggests that those people who are more frequently culturally active are, perhaps unsurprisingly, also likely to be more politically active. However, the idea that cultural omnivores, with wide ranging tastes, are also more likely to get involved in politics is not supported by this analysis of the survey data.

Moving on to social capital, the count of the number of categories of people, based on job status, that respondents know is positively associated with political activity (Beta = 0.077). A one-point change on this variable, which is to say being acquainted with someone from one more category of job status, is associated with around a fifth of a point rise in political activity (B = 0.181). Lastly in the Bourdieusian components of the parsimonious full model, the diversity of respondents’ acquaintances is positively related to political activity (Beta = -0.066, again the variable is inverted so a negative sign indicates a positive relationship). This variable has a short constructed non-integer scale meaning that a full one point change indicates a large difference. Such a change is associated with a two-thirds of a point change on the scale indicating political activity (B = -0.693). Both of these relationships support the idea that being acquainted with more groups of people increases political activity. Crucially, there is no evidence emerging from the current analysis to suggest that knowing people with particular statuses is important, rather it seems to be having a wide set of acquaintances that plays a role.

Finally, and reassuringly in light of the focus of the research, higher perception of the role of privilege in structuring social hierarchies is associated with greater political activity (Beta = 0.060), though not especially strongly. A one point rise on the measure of perception of the importance of privilege in society is associated with approximately a quarter of a point rise in the dependent variable (B = 0.242). This is an interesting finding and suggests that those who see a more important role for privilege in structuring society are apt to undertake more political activity. Again, interpretation at this stage is speculative but the results suggest that perceiving privilege in society motivates a desire to influence politics. Whether this is tendency is stronger amongst those with higher or lower levels of privilege is a question for further analysis but those possibilities suggest quite different motivations.

Sections: Introduction -> Survey Details -> Sample Information -> Survey Items -> Hypotheses and Analyses of Models -> Conclusion

Back to Academic page

 

Advertisements

5 thoughts on “Interim Survey Data Analysis: Hypotheses and Analysis of Models

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s