Sample

This study employed the snowball sampling technique to identify elusive populations while minimizing data collection costs (Biernacki & Waldorf, Reference Biernacki and Waldorf1981). It is based on chain referrals, whereby active participants help identify other participants (Sudman, Reference Sudman1976). Specifically, our data collection process began with a small number of microfinance clients, who then referred us to other microfinance clients. Employees of related microfinance institutions also referred us to other microfinance clients, enabling us to reach a broader audience and achieve greater diversity in the sample. With this method, the final sample comprised 500 completed questionnaires of microfinance clients, 250 of which were from Islamic microfinance institutions and 250 from conventional microfinance institutions.Footnote ²

The questionnaire used in this study can be found in Appendix I. When developing the questionnaire, we considered previous research related to each construct considered. Specifically, Kumari’s (Reference Kumari2021) work informed the measurement of microfinance effectiveness and inspired the poverty alleviation questions. The work of Perry and Morris (Reference Perry and Morris2005) informed the measurement of digital and traditional financial knowledge (DFK and FK). Additionally, Chen, Stanton, Gong, Fang, and Li’s (Reference Chen, Stanton, Gong, Fang and Li2009) work helped us identify how to measure social capital. Furthermore, all the items included in the questionnaire are directly related to our research objective. For instance, we only included items from Kumari’s (Reference Kumari2021) study that were related to microfinance effectiveness.

Five constructs were then calculated using items collected from the survey (see Table 1). The Microfinance Effectiveness (ME) construct comprises four items, which are measured using a five-point Likert scale spreading from ‘Strongly Disagree’ to ‘Strongly Agree’ (see Appendix I for full details). Similarly, the Poverty Alleviation (PA) construct consists of six items. The Digital Financial Knowledge (DFK) construct comprises two items, while the Traditional Financial Knowledge (TFK) construct comprises five items. Finally, the social capital construct consists of five items. Table 1 shows the items used to develop the five constructs.

Table 1.

Constructs with measuring items

To measure Digital Financial Knowledge (DFK), we examined participants’ responses to two specific questions: one concerning credit cards and the other concerning the importance of timely bill payment. As discussed above, credit cards play a key role in digital financial knowledge because they are widely used on digital platforms. Therefore, knowing how to use them is essential in digital financial literacy. For microfinance clients without access to credit systems, understanding credit cards and their associated costs is important for personal financial management (Atkinson & Messy, Reference Atkinson and Messy2012). In the same vein, understanding the importance of timely bill payment is essential for personal financial management in the context of rapid digital technology development. As bill payments are now commonly made through digital tools, being aware of when they are due can prevent microfinance clients from digital penalties (Chhillar et al., Reference Chhillar, Arora and Chawla2024; Koskelainen et al., Reference Koskelainen, Kalmi, Scornavacca and Vartiainen2023; Vieira et al., Reference Vieira, Matheis and Lehnhart2024). Therefore, paying bills on time can demonstrate an efficient use of digital tools, which is directly related to personal financial stability (Ozili, Reference Ozili2018). Regarding microfinance clients in our research sample, most of them are novices when it comes to using mobile phones and digital banking. Therefore, this item allows us to measure their ability to interact with digital finance and to avoid financial losses (Seldal & Nyhus, Reference Seldal and Nyhus2022; Koskelainen et al., Reference Koskelainen, Kalmi, Scornavacca and Vartiainen2023; Lusardi, Reference Lusardi2019).

Table 2 presents the main descriptive statistics of the final sample of 500 participants and shows that 60.6% are male and 39.4% are female. Similarly, there are more young clients than older ones. Furthermore, most participants have an annual income lower than 25,000 Pakistani rupees (PKR), equivalent to around 90 USD per year. Moreover, most of the participants come from rural backgrounds and have an educational attainment level below matriculation. Finally, most participants are married (refer to Table 2 in Appendix I for more details on descriptive statistics of each survey item). In addition, Table 3 shows the correlation matrix for the variables under consideration. Panel 3A-1 relates to traditional financial knowledge, while Panel 3A-2 relates to digital financial knowledge. Overall, correlation coefficients are lower than 0.5, which prevents multicollinearity issues in the estimated model.

Table 2.

Descriptive statistics of participants

Note: The distribution of the main variables collected from the survey is shown in this table. Source: The authors.

Table 3.

A: Correlation matrix. Panel 3A-1: With traditional financial knowledge. Panel 3A-2: With digital financial knowledge

** Note:means that the correlation is significant at the 0.01 level.

Variables

Table 4 defines the variables and their original references. Figure 1 shows how the five constructs under study are measured. To avoid issues of multicollinearity between Financial Knowledge and Digital Financial Knowledge, which have a correlation of 0.962, we construct two separate models. In the first one, only Digital Financial Knowledge (Figure 1) is included while in the second one, only Traditional Financial Knowledge is included (Figure 1A). To save space, the model with Traditional Financial Knowledge (Figure 1A) is presented in Appendix II. Regarding Figure 1, all the constructs are measured through items represented as ME1-ME4, PA1-PA6, DFK1-DFK2, and SC1-SC15, referring to microfinance effectiveness, poverty alleviation, digital financial knowledge, and social capital, respectively. The arrows from each construct to its respective items indicate the strength of their relationship, as measured by factor loadings. These factor loadings are obtained by conducting a structural equation modeling (SEM) analysis using a confirmatory factor analysis (CFA) approach. Higher factor loadings indicate stronger associations between the considered constructs and their measurement items. The error terms, ‘e1’ to ‘e32’, show the measurement errors of the considered items. Figure 1 also shows the relationship among the constructs (the arrows on the left of the figure). This measurement model has been statistically validated, ensuring the reliability and validity of the constructs.

Figure 1.

The constructs’ measurement model with digital financial knowledge

Notes: This figure shows the model to measure the four constructs in the model with Financial Knowledge. ME1 to ME4, PA1 to PA6, DFK1 to DFK2, and SC1 to SC15 correspond to the questions in the survey (see Appendix I for more details). ‘e1’ to ‘e32’ represent measurement errors. The numbers on the arrows on the left side represent factor loadings. The numbers on the arrows on the right side represent the correlations of each pair of variables. Source: From the authors.

Table 4.

Variables

Notes: This table shows the definition of the main variables used in the empirical analysis. ME1 to ME4 refer to the 4 survey items to measure microfinance effectiveness. The same is applied for PA, DFK, FK, and SC. See Table 1 and Appendix I for full details.

Table 5 presents the main descriptive statistics of the constructed variables. The average value of microfinance effectiveness (ME) is 3.8440, out of a maximum of 5. This indicates that participants generally recognize the benefits of microfinance. Similarly, the average of poverty alleviation (PA) of 3.7077 indicates that microfinance has effectively alleviated poverty among survey respondents. Furthermore, the average value of digital financial knowledge is 2.89, which shows that microfinance clients in the sample possess low digital financial knowledge. In addition, an average of social capital of 3.1559 indicates that participants have moderate social connections. Furthermore, the skewness and kurtosis values of the variables are less than 3 and 10, respectively, suggesting that they are not affected by heavy tails or asymmetry (Kline, Reference Kline2011). Additionally, the multicollinearity assessment (Table 6) revealed that the tolerance values for the considered constructs are greater than 0.1, while the values of VIF are lower than 10, indicating the absence of multicollinearity.

Table 5.

Descriptive statistics of the constructs

Notes: This table shows the main descriptive statistics of the five variables. ME refers to Microfinance Effectiveness. PA refers to Poverty Alleviation. DFK refers to Digital Financial Knowledge. FK refers to Traditional Financial Knowledge. SC refers to Social Capital.

Table 6.

Tolerance and variance inflation factor values

Note: Dependent variable = Poverty Alleviation.

Before conducting the SEM estimation, the measurement models need to be validated using Confirmatory Factor Analysis (CFA) (Awang, Reference Awang2015). CFA is used to measure the reliability, validity, and one-dimensionality of the constructs (Awang, Reference Awang2015; Kline, Reference Kline2011).Footnote ³ As shown in Table 7, all the AVE values are between 0.55 and 0.77, which corresponds to the criterion of 0.5. Furthermore, all the Cronbach’s alpha values are above 0.5, confirming the satisfactory internal consistency of the measurement items (Fornell & Larcker, Reference Fornell and Larcker1981; Malhotra, Mukhopadhyay, Liu, & Dash, Reference Malhotra, Mukhopadhyay, Liu and Dash2012). Similarly, the diagonal values (the square root of the Average Variance Extracted) are above the off-diagonal values (Table 8). This result indicates that the measurement items of this construct differ from those of the other constructs in the study. Furthermore, the correlations are less than 0.85 (see Table 8). Therefore, the discriminant validity of the constructs is validated.

Table 7.

The reliability of the constructs

Notes: This table shows the reliability measures of the main constructs considered in empirical analysis. The reliability of one variable is the extent to which an instrument is reliable in measuring the variable (Awang, Reference Awang2015) and is measured through three criteria which are Cronbach’s alpha, Average Variance Extracted (AVE), and Construct Reliability (CR), following previous works such as Fornell and Larcker (Reference Fornell and Larcker1981). Cronbach’s alpha measures the internal coherence of the scale of the variables. A value of Cronbach’s alpha of at least 0.5 shows the reliability of the items in the questionnaire used to calculate the variables. AVE measures the amount of variance that is captured by a construct in relation to the amount of variance due to the measurement error. A value of AVE of at least 0.5 shows that the latent variables from the questionnaire explain at least 50% of the variance of the constructed variables (Awang, Reference Awang2015). CR measures the construct reliability. A value of CR higher than 0.7 is required to achieve construct reliability.

Table 8.

Discriminant validity with digital financial knowledge (with the Fornell and Larcker (Reference Fornell and Larcker1981) method)

Notes: According to Fornell and Larcker (Reference Fornell and Larcker1981). ** Correlation is significant at the 0.01 level (2-tailed), ME = Microfinance Effectiveness, PA = Poverty Alleviation, DFK = Digital Financial Knowledge, SC = Social Capital. Source: Authors’ calculations.

In addition, Table 9 shows the HTMT (Heterotrait-Monotrait Ratio of Correlations) results for assessing discriminant validity. An acceptable level of discriminant validity is with a value lower than 0.9, a criterion that all our variables meet. Tables 7A and 8A (in Appendix III) also demonstrate construct validity when the model incorporates traditional financial knowledge.

Table 9.

Discriminant validity with digital financial knowledge (with the HTMT_0.85 criterion)

Notes: According to the HTMT0.85 criterion. ** Correlation is significant at the 0.01 level (2-tailed), ME = Microfinance Effectiveness, PA = Poverty Alleviation, DFK = Digital Financial Knowledge, SC = Social Capital. Source: Authors’ calculations.

Structural Equation Model (SEM)

Following the validity of the model presented above, Equation (1) illustrates the SEM used to analyze the relationship between microfinance effectiveness (exogenous variable) and poverty alleviation (endogenous variable):

(1)\begin{equation}P{A_i} = {\beta _0} + {\beta _1}M{E_i} + {\varepsilon _i}\end{equation}

where $P{A_i}$ represents the measure of poverty alleviation of participant i. $M{E_i}$ represents the measure of microfinance effectiveness of participant i. ${\varepsilon _i}$ represents the error terms.

To examine the moderating effect of digital and traditional financial knowledge, as well as social capital, we further estimate the equations below:

(2a)\begin{equation}P{A_i} = {\beta _0} + {\beta _1}M{E_i} + {\beta _2}F{K_i} + {\beta _3}M{E_i}DF{K_i} + {\varepsilon _i}\end{equation}

(2b)\begin{equation}P{A_i} = {\beta _0} + {\beta _1}M{E_i} + {\beta _2}DF{K_i} + {\beta _3}M{E_i}F{K_i} + {\varepsilon _i}\end{equation}

(3)\begin{equation}P{A_i} = {\beta _0} + {\beta _1}M{E_i} + {\beta _2}S{C_i} + {\beta _3}M{E_i}S{C_i} + {\varepsilon _i}\end{equation}

where $P{A_i}$ represents the measure of poverty alleviation of participant i. $M{E_i}$ represents the measure of microfinance effectiveness of participant i. $DF{K_i}$ represents the measure of digital financial knowledge of participant i. $F{K_i}{\text{ }}$represents the measure of traditional financial knowledge of participant i. $S{C_i}$ represents the measure of social capital of participant i. ${\varepsilon _i}$ represents the error terms.

Equations (2a), (2b), and (3) include three variables which are digital financial knowledge (DFK), traditional financial knowledge (FK), and social capital (SC), respectively. Importantly, an interactive variable between each of these three variables with microfinance effectiveness (ME) is also included. While coefficient ${\beta _2}$ measures their direct relationship with poverty alleviation, coefficient ${\beta _3}$ allows to capture their moderating role on the link between microfinance effectiveness and poverty reduction. If ${\beta _3}$ is significant and positive, it means that DFK, FK, or SC, has a significantly positive moderating effect. In this case, when microfinance effectiveness increases, an increase in FK, DFK, or SC, will positively influence poverty alleviation. The same is applied for a negative moderating effect. For our research hypothesis to be validated, coefficient ${\beta _3}$ in Equation (2a), related to digital financial knowledge, should be significant and positive.