Chapter 4: What Drives Financial Awareness in Microfinance?
4.4 Results and Discussion
4.4.2 Split-sample Analysis
Table 4.4presents the results of the split-sample analysis, where the main regression is re- peated for different regions in which the MFIs are active and different religions of the re- spondents. We analysis the microfinance borrowers classified into four groups (Africa, Cath- olic Europe, Latin America, and the Middle East) based on two dimensions of cross-cultural variation in the world2: 1. from Traditional to Secular-Rational; and 2. from Survival to Self- Expression. For instance, the clients in the Middle East usually emphasize the importance of traditional value, economic and physical security, while the clients in Europe are more secu- lar-rational and have stronger motivations to pursue self-expression. We also classify the cli- ents into three categories regarding the dominant religion of where they belong to. If the dominant religion takes up less than 40% of the population, the related areas will be identi- fied as a mixture (a great diversity of beliefs). Columns (1)-(2) and (11)-(12) present the re- gressions with variables of regions or religions. It is shown that they are all significant at the 1% level. Respondents who are Islamist or live in the Middle East have much highest odds to correctly remember the interest rates. In comparison, the odds for those from Africa or the countries with multiple religions are lowest.
Columns (3)-(10) and (13)-(18) of Table 4.4, reconfirm that, in Latin America, Middle East, and Christian countries, ๐พ๐ ๐is positively associated to previous access to moneylenders, and negatively associated to previous access to saving service. However, we have found incon- sistent results in other areas. Previous access to moneylenders has no significant influence on ๐พ๐ ๐in the countries with multiple religions. In Africa, previous access to saving service is insignificant. In Europe, previous access to all sources of credit (moneylenders, MFIs and banks) has negative impacts on ๐พ๐ ๐and the impact of previous access to saving service has become positive. One potential explanation for this result is that Europe is a more regulated
2This is a relative scoring method instead of a qualitative description for cultural values. For example, the people in Catholic Europe is more secular than those in Middle East, meanwhile, 72% of respond- ents in the Eurobarometer Survey (2012) described themselves as Christianity. Further details of the definitions and scoring method can be referred to the World Values Survey
financial market than the other regions. Considering the potential convex relationship be- tween interest rate and ๐พ๐ ๐discussed above, if the upper limits of interest rate of all mon- eylenders, MFIs and banks are restricted to very low levels (at the left-hand side of the ver- tex), then the coefficients of previous access to credits will become negative.
Examining the socio-demographic variables once again, we found different results compared to those introduced in subsection 4.4.1. From columns (3)-(10) of Table 4.4, we see that gen- der becomes insignificant in Europe, Middle East, and the countries which are Islamic or have mixed religious. On the other hand, we found significant positive relations between women and the awareness of interest rate in Latin America and Christian countries. In these areas, while men are generally more financially knowledgeable than women, women are more financially cautious than men. In fact, the majority of Christians concentrate in Latin America nowadays. Location and religion are highly correlated. Both Christianity and Islam well recognise marriage and consider a womanโs primary responsibility is to fulfil her role as a wife. But Islam also stresses that a womanโs responsibilities to nurture, educate, and pro- tect her children have taken priority over working and financially support, where women still have the right and are free to work. It may be the reason why women in Christian countries have higher financial awareness than men.
Moreover, age becomes insignificant across all split-samples with multiple imputations after releasing the control of countries. This result is inconsistent with the prior results presented in Table 4.3 as well. With mean imputations, significant results for age are found in columns (4), (8) and (16). Considering the extremely high percentage of missing values in age (40%, see Table 4.1), it is reasonable to guess that the results with multiple imputations are more accurate than those with mean imputation. However, we leave the association between age and financial literacy as unknown in the paper for robustness. Because the accuracy of multi- ple imputations for a discrete and censored variable with very high missing rate is still un- clear. In fact, this issue motivates us to evaluate the imputation performances of different missing data techniques. It will be discussed in details in the next chapter.
What is more, in terms of educational background and living location, the results are also consistent with those shown in Table 3, except for those estimated with the split-sample of Europe, in which both variables have no significant relationship with ๐พ๐ ๐. The results with multiple and mean imputations are the same. Therefore, we can conclude that hypotheses 1.3 and 1.4 are accepted.