Periodo Barroco

Women empowerment is not a one-way traffic; rather it is a never-ending process. The whole process of women empowerment has to be evaluated continuously through the feedback mechanism to maintain a dynamic home statistics. According to Narayana (1998), “Empowerment of women is a process whereby the powerless or disempowered gain a greater share of control of resources and decision-making”. The process of gaining control over self, ideology and resources, which determines power, may also be termed as empowerment.

The indicators of empowerment are the ability to make decisions, participation in household and financial activities, the right of access to property, the dexterity to take shelter in laws and have one’s own power to decide regarding one’s birth and the ability to expand one’s own area. Two of the indexes are usually applied for assessment.

The qualitative indicators of women empowerment are: increase in self-esteem and individual and collective confidence; increase in articulation, knowledge and awareness levels on issues affecting the community at large, and women in particular, such as women’s health, nutrition, reproductive rights, legal rights, literacy etc.; participation in other events related to their lives; increase in the bargaining power of women, as individuals in the home and the community as well as in the collectives of women; women’s decision making power over the kind of work she is doing; her control over her own income and expenditure and whether she is still subservient to male members in the family or not (Raheim & Bolden, 1995).

The quantitative indicators of women empowerment are; demographic trends such as mortality rates, fertility rates, sex ratio, life expectancy at birth and average age of marriage; number of women participating in different development programmes and the participation of women in political processes at the local levels (Mehra, 1997).

The IFAD (International Fund for Agricultural Development) which supported the Tamil Nadu Women’s Development Project in the late 1999’s had done an evaluation and suggested certain measures for the empowerment of women. The Project aimed to bring about the social and economic betterment of women. The empowerment indicators used in developed countries may not fit in developing countries like India where people are trapped in social and cultural bondages. The IFAD model has already been tested in the context of the India, especially in the SHGs under the Mahallir Thittiam. The present study thus adopted the IFAD model of empowerment to evaluate the empowerment indicators.

(a) IFAD Model

IFAD defined empowerment as having the following four main processes:

i) Changes in Women’s Mobility and Social Interaction – This study observed that this type

of change was most likely to occur among women group members when the women involved attained greater self-confidence, had greater respect in the family, played a more assertive role in the domestic sphere, when there was a reduction in domestic violence, when women had greater participation in community affairs and local elections and had more freedom in visit their parents, relatives and friends. The above mentioned nine indices were used to study the changes in women’s mobility and social interactions.

ii) Changes in Women’s Labour` Patterns - This study observed four empowerment

indicators under this. The women involved had better selling and buying skills, independent marketing and better business practices.

iii) Changes in Access to and Control over Resources – The study observed five empowerment indicators under this skill. These included the following: when the women involved played a more assertive role in financial matters, when they gained new skills through training and practice, had better family budgeting, had the right to purchase their own real estates and gain skills for better pricing of their products.

iv) Changes in Intra-Household Decision Making – The study observed the empowerment

indicators as the following: when the women played a more assertive role in the children’s education and health, decisions on recreation, had increased awareness to improve the living standards of the family and community. The above 4 indicators were used to measure the women’s role in maintaining changes in the intra household decision making.

(b) Structural Equation Modelling (SEM)

All the earlier impact studies reviewed in this research used the regression analysis and two stage least square methods. Cohen, Chen, & Dunn (1996) in their study used one of the most conceptual models of impact assessment called the Household Economic Portfolio Model (HHEP). They used the HHEP models to design and test a Structural Equation Model to test the household variables and other variables. This study uses structural equation modelling (SEM), which is a statistical methodology that follows a confirmatory or hypothesis testing approach regarding a proposed causal model generated from theory (Byrne, 2001). Byrne explained two important aspects of the SEM procedure: “(a) that the causal processes under study are represented by a series of structural (i.e., regression) equations, and (b) that these structural relations can be modelled pictorially to enable a clearer conceptualization of the theory under study”. The Structural Equation Model was used to study the impact using the Analysis of Moment Structures or AMOS

(Version 18.0) to perform path analysis. SEM is also known as the Analysis of Covariance Structures or Casual Modelling.

Afrin, Islam, & Ahmed (2008) stated that the method of covariance structure analysis is used to study the implications of the simultaneous regressions primarily at the level of correlations or covariance. The covariance structure model is specified through a simultaneous set of structural linear regressions of a particular variable on other variables. The field of covariance structure analysis actually covers a wide range of topics, including confirmatory factor analysis, path analysis and simultaneous equation and structural equation modelling. The advantage of SEM over the multiple regressions is that it includes more flexible assumptions, correlated independent values and correlated error terms for testing the overall model. AMOS was useful to explore the casual relationships among a set of variables.

The Structural Equation Model was used to study the overall performance of the microcredit. The performance of the micro loans depended on how the participants used their loans in a productive way. If the micro loans were gainfully employed, it increased the income of the household. The total amount of loans used by the members represented the productive base of the households. It was expected to have a significant linear relationship with income and it changed the expenditure, savings pattern, the amount the women spent on their children’s education and their total assets. These changes resulted in a significant linear relationship with better decision making skills of the members, improved health and household conditions and greater empowerment as the end result. Therefore, the total amount of loan taken was expected to have a linear relationship with empowerment. The changes in respondent income, expenditure, savings, amount spent on children’s education, total assets, importance of decision making, health facilities, household conditions and empowerment of the group were considered as dependant or mediating variables. The total

amount of loans of the respondents was considered as an observed exogenous variable and the unobserved exogenous variables or latent variables were e1 to e9. The total variables observed in this model were 19, the number of observed variables was 10, the number of unobserved variables was 9, the number of exogenous variables was 10 and the number of endogenous variables was 9. The figure 4.1 below shows the SEM Model 1 the various relationships of the variables.

Figure 4.1: SEM Model 1

Though the relationships among the constructs were formulated based on the prior literature, it was expected that some model modification would be required in order to reach a best fit model for the data. The SEM model 1 can be tested statistically to determine the extent to which the model is consistent with the data. If a goodness of fit measure is adequate, then the model offers a plausible explanation of the relationships among variable

(Byrne, 2001). Several fit indices were used to make generalizations about the validity of the model by measuring the extent to which the estimated model reproduces the sample covariance matrix (Rakov & Marcoulides, 2000; Nuno, 2008). Model modification was conducted based on examination of the regression weights, Chi-Square and other fit indices. Appendix 13 shows the regression weights of the variables used in SEM model 1. Though the SEM Model 1 shows a satisfactory goodness-of-Fit, this model happened to have a low coefficient estimate for few variables were removed. This is explained in detail under Chapter 5 under structural model assessment.

After screening the variables Figure 4.2 SEM Model 2 depict the measurement model for the independent, exogenous variable (e.g. Total loan amount) and the dependent, endogenous variables (respondent income, expenditure, savings, amount spent on children’s education, total assets, importance of decision making, health facilities, household conditions and empowerment of the group) . The CFA approach confirmed which set of variables were the best indicators for each of the central constructs. It was expected that some of the indicators for each construct would be dropped in order to formulate a best-fit model for each latent construct.

Figure 4.2: SEM Model 2

The Figure 4.2 SEM Model 2 was initially tested and provided an acceptable fit to data, meeting all the fit indexes (see Appendix 14), but several insignificant path which were then removed for model modification to improve the model fit and demonstrate only significant paths of the In conclusion when compared with one another, SEM Model 3 (Figure 4.3) true model proved to be the best fit model for the data. Goodness-of-fit determine if the model being tested should be accepted or rejected. However, overall fit tests do not establish that particular paths within the model are significant. Interpretations of path coefficients should only be reported for good-fit models, as “significant” path coefficients in poor fit models are not meaningful. Good-fitting models produce consistent results on many different indices (Ullman, 2001). However, the comparative fit index (CFI) and root mean square error of approximation (RMSEA) are conceivably the most frequently reported fit indices. It is generally recommended that at least three fit indexes,

choosing at least one from each of the three categories i.e., absolute fits, incremental fits, and parsimonious fits (Nuno , 2008). For the present study, the Chi-Square, RMSEA, GFI, CFI and AGFI were examined and reported as evidence of goodness-of-fit.

The model Chi-Square is the most common fit test in SEM. A significant Chi- Square (P<0.5) translates to a bad fit, demonstrating a significant difference. However, Chi- Square used in SEM must be interpreted with caution since it is extremely sensitive to large sample sizes (Byrne 2001).

Therefore, it is recommended that Chi-Square statistic in a study should not be the only means of assessing the model fit. Root mean square error or approximation (RMSEA) like Chi-square is a badness-of-fit measure but also takes into account of parsimony. By convention, there is good model fit if RMSEA is less than or equal to .05, and there is adequate fit, accordingly an “accepted fit” if RMSEA is less than or equal to .08 (Nuno, 2008). Goodness-of-fit index (GFI) measures the relative amount of variances and covariances that are accounted for by the model. The index ranges from zero to 1.00 with values closest to 1.00 (e.g., values greater than .90) as being indicative of good fit. (Byrne 2001). Comparative fit index (CFI) depends on the average size of the correlations in the data. The values of CFI range from zero to 1.00, where scores > .90 indicate an acceptable fit (Byrne 2001).

Figure 4.3: SEM Model 3

SEM techniques were performed using Amos version 18.0, employing Maximum Likelihood (ML) estimation method. ML is the most common method of estimating the best fitting parameters for SEM (Schumacker & Lomaz, 2004). The study constructs were individually tested using a factor analytic model approach to statistically test how and the extent to which the observed variables were linked to their underlying latent factors. The measurement models assessed the overall factorial structures utilizing fit indexes, and if necessary, the misspecifications which resulted in poor measurement are removed. The CFA approach provided valuable insight for model modification to achieve a better data-to model fit, and helped to construct the test the structural model. The model fit summary is explained detail in Chapter 5 under the heading Structural Model Assessment.