We tried to include as many explanatory variables as possible. Some had to be eliminated due to severe missing data problem, as referred to above. The final model selected was the following:
owe_money = F( income_pct employment_status education_level female age agesq country)
The variables are from among those detailed in Table 1 in the Appendix. The agesq (=age^2) term, common in the literature, is included to avoid unrealistic monotonic age effects.
The results are presented below:
xtprobit owe_money income_pct employment_status education_level female age agesq country , re difficult allbaselevels
Wonder how to write the most effective application letter for college? Visit Pro-Papers academic application letter writing service and find some really useful hints for writing.
Number of observations - This is the number of observations that were used in the analysis. This number is a bit smaller than the total number of observations (30,014) in the data set, due to missing values for any of the variables used in the logistic regression. (Stata uses a list wise deletion by default, which means that if there is a missing value for any variable in the regression, the entire case will be excluded from the analysis).
The overall significance of the regression is shown by the Wald test statistic, and the Prob > chi2 - This is the probability of obtaining the chi-square statistic given that the null hypothesis is true. In other words, this is the probability of obtaining this chi-square statistic (1257.25) if there is in fact no effect of the independent variables, taken together, on the dependent variable. This is, of course, the p-value, which is compared to a critical value, perhaps .05 or .01 to determine if the overall model is statistically significant. In this case, the p-value is reported as 0, so the overall model IS statistically significant
Further, the model gives us highly significant estimates of coefficients for almost all the independent variables, the only exception being income_pct.
The chi-squared test on the estimated rho indicates that the true rho is NOT zero, nor does it seem likely to be 1, since the confidence interval does not include 1. Hence, while, RE is superior to a Pooled Regression, the estimated b's will be likely to be somewhat biased, due to unobserved heterogeneity.
Coefficients: As explained earlier, the Probit coefficients are a bit difficult to interpret - but they do indicate the directions and size of effect of each variable on the probability of the independent variable. Thus income (weakly) & education (strongly) have a positive effect on the propensity to borrow.
The other independent vars. have negative effects:
female=1, male=0, so the negative sign means that men are less likely to be borrowers.
Similarly, age coefficient negative implies older people less likely to borrow.
The other effects are quiet are difficult to interpret, since they refer to factor variables, i.e., categories, such as employment & country. It is easier to see the meaning if we obtain explicit parameters for the factor levels. This is done in the following modified regression:
xtprobit owe_money income_pct i. employment_status i. education_level i.female age agesq i.country , re difficult
Here i.employment_status, i.education_level, i.female, and i.country are Factor variables- one dummy variable is calculated by Stata for each level of the variable, except one, the base level. So there are 6 dummy variables for the 7 countries (country=12 (Germany) is the base level), 1 for female (gender), 3 for the 4 categories of education_level, 4 for the 5 categories of employment_status
To interpret the coefficients, we need to know the value codes used, and they are given in the Table below.
Hence, we can say that for given values all other independent variables, as compared to a retired person, the employed are somewhat likely to borrow more, the unemployed a little less so, the disabled much more, and homemakers less.
Similarly, for given values all other independent variables, compared to those with no education, all the other categories are more likely to borrow, with the highest category most likely. Again, for given values all other independent variables, Females (code=1) are less likely to borrow than males (base). Finally, among the countries, compared to individuals in Germany, for given values all other independent variables, those in Sweden (13) & France (17) are more likely to borrow, while those in Netherlands, Spain & Italy are less likely.
Categorical variables Codes
As mentioned earlier, the regression coefficients, while they establish the direction of the relationship, the Probit regression coefficients cannot be interpreted in terms of actual changes in probability of borrowing. This can be seen from the Marginal Probabilities. These are the probabilities of borrowing calculated for each level of a categorical variable, given all other variables at their mean levels. They are given below.
margins country, predict(pu0) atmeans grand
So, as indicated by the regression coefficients, compared to individuals in Germany (who have an average 5% probability of being a borrower), those in Sweden(13) with 22% & France (17) with 17% are more likely to borrow, while those in Netherlands, Spain & Italy have much lower probabilities of borrowing.
margins employment_status, predict(pu0) atmeans grand
Here we see that the Retired on average have a 6.4% probability of borrowing, the Employed, 8.6% etc. Surprisingly perhaps, the highest probability is for the Disabled. Homemakers, not surprisingly, have the lowest probability
margins female, predict(pu0) atmeans grand
So, as expected from the discussion earlier, Females (code=1) are less likely to borrow than males (0),the respective probabilities being 6.3% & 7.7%
margins education_level, predict(pu0) atmeans grand
Similarly, for given values all other independent variables, compared to those with no education (code=0), all the other categories are more likely to borrow, with the highest education category having the highest probability- 8.33%.
The above probabilities are computed for the entire sample of all 3 waves. It would be interesting to see how the patterns may have changed over the waves. For the country-level probabilities, and employment categories, this is shown by the following graphs:
Figure 1 Country-wise probabilities of borrowing across Waves
Figure 2 Employment Status-wise probabilities of borrowing across Waves
It is apparent from the literature that has been studied as well as the data analysis that one specific trend that is valid globally is that consumer credit is more freely available now as well as there is a greater utilization of consumer credit over the other conventional financial products like mortgages. It is also very apparent that the users of consumer credit are growing younger, i.e., the average age of the borrower is steadily decreasing, and it is found that more and more educated people are approaching financial institutions for availing of such credit. When compared to older economies like Japan and China, utilization of consumer credit is far higher in the European economies. Older people seem to be not using as much consumer credit as the younger population. In addition, the uneducated are shying away from consumer credit. Several factors could be attributed to this, which however, is beyond the scope of this study. One surprising finding is that, people with disabilities are using higher amount of consumer credit than others. Moreover, men are using lesser consumer credit when compared to women.
Looking for some amazing hooks for essays that could come in handy? Seek no more, Pro-Papers presents a unique and amazing blog post about the greatest ever hooks for essays writing!
From a marketing perspective, it would be logical for the providers of consumer credit to target women, educated youth up to and beyond college levels without any disabilities. Another category would be the younger population with education as well as disabilities. The third possible category is people with disabilities. These three categories of people seem to have consumed greater amount of credit over the last two decades or so and it can be safely concluded that there would be sufficient growth in these sectors.
A very interesting phenomenon is that the senior citizens seem to be shying away from consumer credit, possibly because of lower income levels; however, there is no sufficient data to conclude on this. Consumer credit has been an interesting product that has found and made deeper in-roads into the consumer population across the European Union and tends to now compete with other mortgage products available and it is obvious that consumer credit has found greater utilization in the European economies in the last two and half decades.
Leave a Reply
Your email address will not be published / Required fields are marked *