All pages

Chapter 2: Explaining Social Trust

The study of trust is bedevilled by the problem of cause and effect. Do people become more trusting as a result of close and sustained interaction with others in voluntary organisations? Or is it, rather, that trusting people join voluntary associations and get involved with their community, leaving distrusting ones at home to watch television? Do people develop higher levels of trust because life has been kind to them, or is life kind to them because they are trusting? Many commentators have pointed out the severe chicken-and-egg problem associated with most theories and empirical findings about trust, and we are unable to make much progress with the problem here. But it is worth making two important, if preliminary, observations about cause and effect:

  1. First, we look for close associations between a varied set of independent variables and our measure of social trust. If we find such associations, then we can begin to worry about which is cause and which is effect. If we do not find close associations, then there are no problems of cause and effect to ponder on in the first place.
  2. Second, there is no general rule about how to determine the direction of causal relations, at least when one is dependent upon cross-sectional survey data. Each particular combination of figures has to be examined independently to see what causal relations are theoretically plausible and implausible. This is not a statistical matter as much as one for the social and political imagination.
Page 1

Two possible origins of social trust

In this chapter we are primarily interested in investigating which of the two theories under discussion can best explain social trust.

The first is the classical theory that participation in voluntary organisations creates "the habits of the heart" that underpin support for democratic politics and the ability and inclination to become involved in politics. Voluntary organisations, it is said, teach people respect and understanding for others, even though the others may belong to a different social background (social class, ethnic group, gender, age, religion). Voluntary organisations also draw people into the community, engage them socially, and teach the art of organising, co-operating and compromising with others in order to achieve collective goals. In short, voluntary associations are said to encourage trust, reciprocity, and political understanding and skills.

Figure 2.1. Voluntary organisation theory

The second is the individual-oriented theory, which we have termed the Success and Well-being theory. This theory states that personal experiences in adult life influence how one interprets other persons to be. It is more risky for the poor to trust others because even small losses can have big consequences. The implication is that those who are well off are more likely to trust other people than poor people are.

Figure 2.2. Success and well-being theory

Tests of these two theories can be designed in many different ways. In this hands-on example we will include all respondents in all the countries covered by the ESS in the calculations. It is possible that the results would have been a bit different if the analysis had been conducted in relation to one nation at a time. When one uses all respondents, one misses the possibility of discovering intra-country differences.

Note that you are expected to follow the procedures described below:

Page 2

Voluntary organisation theory

In the following exercises, you will be asked to perform operations that are not possible to do online. This is why you should download the dataset "Download: Trust" to your own computer. You could download the data in any of the available formats, but you must be aware that all syntax examples will be given with reference to SPSS.

  1. Download the dataset Download: Trust and save the dataset to your disk.

From Table 1.1 we saw that the voluntary organisation theory has been measured using membership of different associations/organisations. The ESS survey asked about membership of twelve different types of organisations. (Please note that the voluntary organisation questions were not asked in Switzerland and in the Czech Republic.) By adding the information from these twelve variables into one variable, we get a more robust measure of the involvement in the civic society.

Figure 2.3. Operationalisation of the voluntary organisation theory

  1. Create a variable that summarises the twelve original variables with information about membership of different organisations. SPSS

    You could create a simple additive index by running the syntax below. The index will vary from 0 (not a member in any voluntary organisation) to 12 (member of all types of organisations).

    Compute vol_org =SUM(sptcmmb, cltommb, trummb, prfommb, cnsommb, hmnommb, epaommb, rlgommb, prtymmb, setommb, sclcmmb, othvmmb).
    VARIABLE LABELS vol_org "Additive index: Membership in 12 types of voluntary organisations".
    FORMATS vol_org (F2.0).
    EXECUTE.
  1. Examine the distribution of the computed variable - i.e. how many people score 0, 1, 2, 3 …up to 12. What is the shape of the distribution? Does the shape make sense to you given what you know about the likely membership patterns of the people you know? Does the shape of the distribution have any implications for the regression analysis that follows? SPSS

    To examine the distribution, you could run a frequency analysis and draw a bar graph of the variable vol_org.

    FREQUENCIES
    VARIABLES=vol_org
    /ORDER= ANALYSIS.
    GRAPH
    /BAR(SIMPLE)=COUNT BY vol_org.

    The distribution is highly skewed. Approximately 40% of respondents are not members of any organisation, and fewer than 25% are members of three or more organisations. Because there are so few respondents who are members of several organisations, these respondents might have a large impact in the following regression (outliers). In short, this variable is not particularly suited for regression.

There are at least two reasons for being somewhat sceptical about using the variable computed in the exercise above as a measure for civic involvement.

Firstly, it is a weakness that the level of personal engagement in the organisation is not accounted for. A passive member is assigned the same value as a more committed member.

Secondly, the assumption that the individual's involvement increases in a linear way through the addition of each new membership is doubtful. Is it likely that a person playing in a brass band, singing in a religious choir and playing in a chess club is three times as involved as a person who is only a member of a political party?

The first problem could be met by expanding the index to include variables that could say something about the level of involvement, for example if the respondent has done voluntary work for the organisation. The second problem could be met by emphasising the first couple of memberships more than later ones. If a respondent has three memberships, it is reasonable to expect that a new membership will have less impact on his civic involvement than the first and the second membership had. Practically, this can be done by transforming the additive scale into a logarithmic one, to reduce the effect of large values on the index.

  1. Create a variable called "Voluntary organisations: Logarithm of membership and involvement index" according to the description above. For each of the twelve types of organisations you should use three variables: "member", "participated" and "voluntary work". In total this makes 36 variables. The first step is to add these together, and we get an additive index varying from 0 to 36. The second step is to take the logarithm of this index. SPSS

    The first step is to make the additive index:

    COMPUTE org=SUM(sptcmmb, sptcptp, sptcvw, cltommb, cltoptp, cltovw, trummb, truptp, truvw, prfommb, prfoptp, prfovw, cnsommb, cnsoptp, cnsovw, hmnommb, hmnoptp, hmnovw, epaommb, epaoptp, epaovw, rlgommb, rlgoptp, rlgovw, prtymmb, prtyptp, prtyvw, setommb, setoptp, setovw, sclcmmb, sclcptp, sclcvw, othvmmb, othvptp, othvvw).
    EXECUTE.

    The second step is to take the logarithm of this index. Please note that it is necessary to add 1 to the additive index before computing the logarithm. 0 is a valid value indicating that the respondent is not in any way involved in organisational work. But the logarithm of 0 is nonexistent, and the logarithm of 1 is 0, and that is what we need:

    COMPUTE ln_org = LN((org+1)).
    VARIABLE LABELS ln_org "Voluntary organisations: Logarithm of membership and involvement index".
    EXECUTE.
Page 3

Success and well-being theory

In Chapter One, see Table 1.1, we associated five rather different variables into this theory. We could of course recode these variables and create an index, but we will use another approach. First we will identify which of these variables are most strongly connected to social trust, and then we will use this result to design a factor analysis. By running a factor analysis, we have the opportunity to save each respondent's score on the first factor. Because all variables are expected to express the underlying theoretical dimension (which is common to all four variables), we can use this score instead of the original variables. In this way we can conveniently reduce our four variables to one, and therefore make the analysis that follows both easier and quicker. By taking a single factor that underlies all four questions in the original questionnaire we can also minimise and cancel out the problems and errors ("noise") associated with any one question - that is, we can maximise the reliability and validity of our four questions.

Figure 2.4. Operationalisation of the success and well-being theory

  1. Identify which of the variables "stflife. How satisfied with life as a whole", "happy. How happy are you", "unemp3m. Ever unemployed and seeking work for…", "income. Households approximate Monthly net income…" and "hincfel. Feeling about household's income…" correlates most strongly with the trust variable ("ppltrst. Most people try to take advantage of you…"). Please remember to use the combined weight variable. SPSS
    WEIGHT BY COMB_WEIGHT.
    CORRELATIONS
    /VARIABLES=ppltrst stflife happy uemp3m income hincfel
    /PRINT=TWOTAIL NOSIG
    /MISSING=PAIRWISE.
    WEIGHT OFF.
  1. Exclude the variable with the lowest coefficient, and include the remaining four variables in a factor analysis. (Please remember not to include the trust variable!) Save the factor scores. The first factor extracts as much as possible of what these variables might have in common. Our assumption is that each respondent's score on this factor could be used as a measure of the degree of success and well-being felt by the respondent. SPSS
    FACTOR
    /VARIABLES stflife happy hincfel income
    /MISSING LISTWISE /ANALYSIS stflife happy hincfel income
    /PRINT INITIAL EXTRACTION
    /CRITERIA MINEIGEN(1) ITERATE(25)
    /EXTRACTION PC
    /ROTATION NOROTATE
    /SAVE REG(ALL)
    /METHOD=CORRELATION.
    RENAME VARIABLES FAC1_1 = fac_suc.
    EXECUTE.
    VARIABLE LABELS fac_suc 'Success and well-being: Regr factor score'.
    EXECUTE.
Page 4

The two theories combined

The following step is undertaken to assess which of the two theories, as operationalised here, is the most powerful determinant of social trust. We can do this by means of ordinary least square (OLS) regression. By including the two constructed variables as independent variables, and the trust variable as the dependent variable, we can measure the effect.

  1. Perform the regression analysis. Please remember to use the factor score variable and the logarithm of membership and involvement index variable. You must also remember to use the combined weight. Interpret the regression equation (y=a+b1x1+b2x2). SPSS
    REGRESSION
    /MISSING LISTWISE
    /STATISTICS COEFF OUTS R ANOVA
    /CRITERIA=PIN(.05) POUT(.10)
    /NOORIGIN
    /DEPENDENT ppltrst
    /METHOD=ENTER ln_org fac_suc.
  1. Which of the theories seems to explain most of the variation on the trust variable? Solution

    Compare the standardised regression coefficients, beta. According to the beta, Success and well-being theory is a far more powerful determinant of social trust than the voluntary organisation theory. But even though the beta is a standardised measure, you should use it with caution when the variables are not following the same scale. To get an extra argument for your conclusion, you could perform two bivariate regression analyses, one for each of the theories, and compare the amount of explained variance, R square.

The results also reveal that these two theories are far from the only sources of social trust. The overall explanatory power of the model is about 15 %. This is statistically significant (unlikely to have occurred purely by chance), but substantively it explains only a small part of the social basis of trust. But considering the type of data used here (categorical dependent variable with ten categories, data from many different countries), 15 % is a good result.

  1. Pick four countries. Perform a regression analysis similar to the one above for each of these countries. It is only necessary to use the design weight. SPSS

    Start by switching on the weight. Then select a country, as the syntax example does. Select your own country or, if you do not live in an ESS country, select Norway as in the following example (If you do not know which code each country has, go to Variable view in SPSS, and to the column "Values". Click on the cell for the "Country" variable. Then you will get a list of codes and countries.)

    WEIGHT BY dweight.
    USE ALL.
    COMPUTE filter_$=(Country = 18).
    VARIABLE LABEL filter_$ 'Country = 18 (FILTER)'.
    VALUE LABELS filter_$ 0 'Not Selected' 1 'Selected'.
    FORMAT filter_$ (f1.0).
    FILTER BY filter_$.
    EXECUTE .
     
    REGRESSION
    /MISSING LISTWISE
    /STATISTICS COEFF OUTS R ANOVA
    /CRITERIA=PIN(.05) POUT(.10)
    /NOORIGIN
    /DEPENDENT ppltrst
    /METHOD=ENTER ln_org fac_suc .
    FILTER OFF.
    USE ALL.
    EXECUTE.
    WEIGHT OFF.
    1. Do you find any differences between the four countries? If there is little difference between the countries, do you think you have found a universal social pattern? If there are differences between the countries, what sorts of factors might explain them?
    2. Are the results statistically significant? Are they substantively meaningful?
    3. Is the individual theory still a more powerful predictor than the social one?
    4. This question can't be solved using NSDstat: Experiment with different regression models, by putting the success and well-being variable into the equation first, followed by the voluntary organisation theory, and then switching the order. Use forward regression and backward regression. Does it make any difference to the results? If so, why do you think it makes a difference?
    5. What does the regression coefficient tell us about cause-and-effect relations?
    6. What do you think might be a better predictor of social trust?
Go to next chapter >>
Page 5