This study uses cross-sectional, nationwide data from Kenya’s 2021 FinAccess Household Survey, comprising 22,024 households. Of these, 6,134 have access to mobile money, and 15,890 do not. Additionally, we have demographic and socioeconomic characteristics for each household, including age, gender, mobile phone ownership, educational level, marital status, and area of residence. Table 1. summarizes the variables used in this study.

Table2. displays summary statistics of the paper. Households in the treatment group spent more on rental purposes than the control group by 892.5 Kenyan Shilling (14.61 USD), with high statistical significance (often p < 0.01). Similarly, the treatment group’s food spending surpassed the control group's by approximately 308.2 Kenyan Shilling (2.38 USD), indicating statistical significance. Moreover, households in the treatment group allocated approximately 142.6 Kenyan Shillings (1.10 USD) more to cellphone expenditure than those in the control group, representing a statistically significant difference. There was also a substantial difference in transportation spending, with the treatment group investing approximately 191.1 Kenyan Shilling (1.48 USD) more than the control group. Additionally, the control group had a 9% higher proportion of females than the treatment group, which was statistically significant. The treatment group had a higher percentage of educated individuals (97%) than the control group (76%), indicating a substantial gap of 21 %. Although the control group had a slightly higher proportion of single individuals (2 %), this difference was statistically significant. The treatment group owned a phone (97%) compared with the control group (74%), suggesting a noticeable contrast ( Table2. ).

Moreover, the treatment group had a 21% lower proportion of rural households than the control group, with a statistically significant difference. Household age of the treatment group was approximately 3.85 years lower than that of the control group, demonstrating a statistically significant variance. A large sample size enhances the reliability of statistical analyses. The treatment group with access to mobile money exhibited significantly different expenditure patterns and demographic characteristics from the control group.

Access to mobile money is non-random, meaning that comparisons between those with access and those without access could be influenced by self-selection. To address this potential selection bias, we employ a matching method, which pairs treated individuals (those with access to mobile money) with untreated individuals (those without access) who share similar pre-treatment attributes (Bari et al., 2024) (West et al., 2014). This approach assumes conditional independence (CIA), meaning that, after controlling for pre-treatment covariates (i.e., X), the treatment assignment (mobile money access) is as good as random. This assumption allows us to estimate treatment effects by comparing outcomes between matched individuals who have similar characteristics, regardless of whether they have access to mobile money.

The equation for the Average Treatment Effect on the Treated (ATET) is as follows:

ATET(x)=E [Y₁ |D=1, X=x] - E [Y₀ |D=0, X=x] (1)

In the given scenario, Y denotes four expenditures: food, rent, transportation, and cellphone expenses (results). X represents the set of pretreatment covariates, and D is the treatment dummy variable that characterizes a household’s use of mobile money. D = 1 means that a household has access to mobile money, and D = 0 means that it does not have. E [Y1 |D=1, X=x] refers to the expenditure of treated households, whereas E [Y0 |D=0, X=x] refers to the expected expenditure for the best untreated match.

The equation used to estimate the ATET under the propensity score P(x) is as follows:

ATET = E[Y1|D=1,P(x)] − E[Y0|D=0,P(x)] (2)

Table 1. presents the variables used in this study. The dependent variables encompass expenditures on food, cell phone usage, rent, and transportation, while the independent variable is the utilization of mobile money.

Table 3 reports Caliper and Kernel Matching results of the impact of access to mobile money. The result consistently shows that access to mobile money significantly increases spending on food, rent, cellphone, and transportation, at a 1% significance. According to Caliper and Kernel Matching, the food expenditure increased by 207.47 and 146.18 Kenyan Shillings (1.60 and 1.13 USD), respectively. According to Caliper and Kernel Matching, the rent expenditure increases by 653.17 Kenyan Shilling (5.04 USD) and 598.59 (4.62 USD) Kenyan Shilling, respectively. Moreover, according to Caliper and Kernel Matching, cellphone spending increased by 112.67 Kenyan Shillings (0.87 USD) and 110.81 Kenyan Shillings (0.86 USD). Further, transport spending increases by 134.43 Kenyan Shillings (1.04 USD) and 149.83 Kenyan Shillings (1.16 USD), according to Caliper and Kernel Matching.

We employed the IPWRA estimation to evaluate the robustness of the results. The results in Table 4 are consistent with our primary findings, demonstrating that financial inclusion contributes to an overall increase in expenditures. Table 4 summarizes the influence of mobile money on food, cellphone, rent, and transport expenses, as identified through the IPWRA approach. The ATET analysis indicates that mobile money positively impacts these expenses, as noteworthy p-values indicate.

Balancing the covariates in PSM is essential to accurately estimating the impact. The matching process successfully equalized the covariates between the two groups. Any significant biases present before matching were significantly reduced, as evidenced by the substantial decrease in bias percentages and nonsignificant p-values after matching. The findings demonstrated that the treatment and control groups are now more comparable, thus enhancing the validity of the subsequent analyses.

The propensity score matching method requires assessing the balance between the covariates used in the estimation. Table 5 summarizes the balancing check of the treatment and control groups before and after matching. Table 5 shows that all the matching covariates were statistically different between the control and treatment groups before the match. After matching, the mean values of the covariates were not significantly different. Figure 1shows the propensity score distributions of the covariates. The propensity score indicated a slight overlap between the two groups. Before matching, the graphs showed that the distributions of the covariates were inconsistent between the control and treated groups. However, after the matching process, the distribution became more uniform.

The results demonstrated that access to mobile money services leads to an increase in spending on food, cellphone, rent and transport purposes respectively. These findings suggest that access to mobile money can promote household level expenditures. Our findings align with those of (Suri & Jack, 2016), who investigated the enduring effects of mobile money on household consumption in Kenya using M-Pesa. Their research revealed that mobile money positively affects Kenyan household consumption patterns and reduces poverty rates.

Moreover, our study corroborates the findings of (Djahini-Afawoubo et al., 2023), who demonstrate that mobile money effectively alleviates poverty among low-income individuals in developing countries. Their study suggests that mobile money significantly reduces multidimensional poverty, benefiting rural residents, women, and those with lower literacy levels. Our findings align with those of (Suri & Jack, 2016), who investigated the enduring effects of mobile money on household consumption in Kenya using M-Pesa. Their research revealed that mobile money positively affects Kenyan household consumption patterns and reduces poverty rates.

However, while previous research has focused on the impact of money on financial issues, this study explores the impact of mobile money on household expenditure. Our findings are consistent with a survey conducted by (Dube & Chummun, 2019) in Zimbabwe, which found that mobile money positively influenced household livelihoods. This indicates that mobile money can improve food security, enabling households to maintain consistent consumption patterns and avoid interruptions in purchasing critical items. Furthermore, this study underscores the importance of boosting financial literacy to empower households to handle their finances through mobile monetary platforms. From a policy perspective, these findings highlight the need to promote mobile money as an integral part of holistic financial inclusion. Governments and financial institutions must continue to enhance the mobile money infrastructure, particularly in underserved areas.

Additionally, integrating financial literacy programs with mobile money services has the potential to mitigate potential risks and maximize the benefits for households. The study’s limitations stem from possible biases in self-reported data, difficulties establishing causation, and the limited applicability of the findings beyond Kenya. A more extensive examination of the broader economic and policy landscape and behavioral shifts linked to mobile money usage could also enhance this study. Therefore, future research should explore the prolonged impact of mobile money on household expenditure and its effect on other facets of household well-being, such as education and health. Overcoming these challenges may necessitate alternative methodological strategies, such as randomized controlled trials, to establish causation definitively.