Measuring and Optimizing for Financial Health

By Jennifer Chan Yih

At Tala we are constantly evaluating new ways to understand our impact on our customers. Most recently, Tala Data Scientist Jennifer Chan Yih developed a new methodology to measure subjective financial health.

Measuring Financial Health for the Benefit of Our Users

At Tala, our mission is to enable and accelerate financial health. To track our progress towards our mission, we needed a way to quantify, measure, and track financial health among our customers.

Beyond measuring our progress towards our mission, how will measuring financial health help our users? Currently, we track objective measures of financial health, such as loan repayment behavior across time. However, we wanted a subjective measure of financial health that captures our users’ understanding of their financial health － of their abilities to manage their day-to-day expenses, prepare for emergencies, and plan for their futures.

By developing scales for quantifying and measuring financial health, we can more holistically track how any changes to algorithms, policy, or product features shift the financial health of our users. This means we can more effectively optimize our business and our products for financial health.

So I set out on a journey to develop long- and short-form scales for measuring self-reported or subjective financial health. Previously as an academic researcher and psychological scientist, I have used a similar approach to create and validate scales for measuring self-reported emotional experience (Kirby, Morrow, & Yih, 2014; Sung & Yih, 2015; Sung & Yih, 2019) and related cognitions and behaviors (Yih, Kirby, & Smith, 2019; Yih et al., 2019). Importantly, each of my past experiences focused on scale development in a single cultural context, whereas this initiative on financial health at Tala spanned three of our markets (Kenya, Mexico, and the Philippines). Through working on this current initiative and analyzing the empirical data, I have definitely gained a deeper appreciation for fine-tuning scales across cultures, countries, and continents.

Defining Financial Health

In line with research conducted by the Financial Health Network and FSD Kenya, at Tala we use a three-part definition of financial health:

Use: ability to manage daily finances
Protect: ability to prepare for and to cope with financial emergencies and shocks
Grow: ability to plan, save, and invest for the future

Notably, these definitions focus on specific behaviors and capacities. In addition to measuring these capacities, I wanted to capture and track the emotional aspect of financial health:

Happiness: positive sentiment towards one’s financial life

Below I describe how I developed reliable scales for measuring each of these constructs.

Developing Scales for Measuring Financial Health

I developed scales to reliably measure each capacity (Use, Protect, Grow) and also financial happiness (Happiness). My goal was to develop internally consistent and reliable scales that measured my constructs of interest in an efficient way. In other words, I wanted to use a few items to stably measure the different components of financial health. Towards this goal, I needed to create a repository of potential items and then, using theory and empirical data, whittle that list down to a small set of items that reflect the intended construct and are correlated to each other.

For each construct (Use, Protect, Grow, and Happiness), I used the following process:

Using a review of the existing literature as well as previous research conducted by Tala, I identified several individual items hypothesized to measure the intended construct.
Design and launch a survey to collect responses to each item, using an apriori power analysis to determine minimum sample size for your survey.
Compute the descriptive statistics for each item, and analyze the internal consistency and reliability using Cronbach’s alpha.
Identify a set of items that holistically represents your construct of interest, while avoiding conceptual redundancy, that has a Cronbach’s alpha coefficient ≥ .70 (which is the standard in academic social science research).
To establish the validity of your measures and explore how well they measure what they are theorized to represent, correlate measures of subjective financial health with more objective indicators of financial health, such as behaviors around loan repayment.
If you want to measure financial health across multiple cultural contexts, consider launching the same survey to these other segments/countries and testing if your results replicate (i.e. that the set of items remains internally consistent and reliable as indicated by the Cronbach’s alpha coefficient). If you want to create a global scale, you may need to reselect your items to create a set that is more universally reliable across cultures.

Use Scale

We recommend measuring the ability to manage daily finances using the following items (with the items for the shortened version of the scale in bold):

In the past month, I paid all my expenses on time.
In the past month, I used a budget to manage my finances.
In general, compared to other people my age, I am very good at managing my day-to-day finances.
I feel that I can keep up with daily expenses.
I have all the financial tools and products that I need.

We use a 5-point Likert scale to record responses for each item (pictured above), as well as for all items detailed below for the other constructs. Also, we typically insert the specific timeframe such that “In the past month” becomes “In May 2020” for instance. To compute the Use score, take the mean of the items (3 items in the shortened version).

Protect Scale

Next, we measure the ability to prepare for and to cope with financial emergencies and shocks using the following items (with the items for the shortened version of the scale in bold):

In the past month, I had enough money to pay for an unexpected expense up to KSh 4,000, if needed.
In general, compared to other people my age, I am very financially prepared for unexpected situations.
In the past month, I had enough money to pay for an unexpected expense over KSh 10,000, if needed.
For next month, I could manage my expenses with only my savings and assets.
I feel prepared to handle emergencies and unexpected expenses up to KSh 4,000.

To create the Protect score, take the mean of the items (3 items in the shortened version).

Grow Scale

We recommend measuring the third definition of financial health － the ability to plan, save, and invest for the future using the following items (with the items for the shortened version of the scale in bold):

I feel that I can plan for opportunities in the future, such as buying land or starting a business.
In general, compared to other people my age, I am able to pursue my financial and life aspirations.
In general, compared to other people my age, my financial health is excellent.
In the past month, I set aside money for my individual savings.
In the past month, I saved regularly by putting money aside every week.

To compute the Grow score, again calculate the mean of the items (3 items in the shortened version).

Happiness Scale

Finally, we recommend measuring positive sentiment towards one’s financial life using the following items:

I feel happy about my current financial situation.
I am financially well-off and stable.
I am stressed about my financial future. [reverse-scored]
I am worried about my financial future. [reverse-scored]
I feel burdened by the amount of loans/debt that I have right now. [reverse-scored]

Here I denote the three items that should be reverse-scored, such that the Likert scale is reversed (i.e. 1 = Strongly agree and 5 = Strongly disagree). As with the other constructs, to create the Happiness score, calculate the mean of the items. Given variability in the item reliabilities across cultures, we recommend using a 5-item version of this scale.

Overall Financial Health Metric

Dashboards can be setup to track each construct of interest, but sometimes, organizations may prefer monitoring a single metric, perhaps for the sake of simplicity and to streamline the task of optimization. We do the same at Tala, taking the mean across our scores for each financial health construct (Use, Protect, and Grow) to create an overall financial health metric. As a reminder, Happiness is not formally included in our definition of financial health; rather, Tala is interested in tracking financial happiness to understand user sentiment towards personal financial health.

Across all countries, this overall financial health score is strongly correlated with each subscore (Pearson’s r > .75). See below for a correlation matrix from the scores derived from our users in the Philippines; the first row represents the overall metric, with the 3- or 5- item versions of each construct denoted in the subsequent rows of the matrix:

Using a single comprehensive metric, we have the ability to not only concisely track changes in financial health, but we also have the power to zoom in at any time and to decompose why the metric may be trending in a given direction for a specific individual or group.

For example, we may want to observe how financial health is impacted by a change in policy for our core credit product. In this hypothetical situation, let us assume that we will use an experiment to monitor and to formally test the effect of the new policy. In this experiment, one key metric will be subjective financial health. After launching the experiment to a segment of users who receive the treatment (new policy), we observe that our overall financial health metric significantly increases when comparing this metric before vs. after the application of the new policy. Yay, that is promising news! However, we may want to dig deeper and understand why financial health improved. Was it due to an increase in the ability to manage daily finances, or in the ability to prepare for emergencies? Both? Or neither － perhaps the mechanism was the improved ability to plan for the future, or increased financial happiness? It is easy to decompose the overall metric into its subscales (Use, Protect, and Grow) to compute each separate score and to observe fluctuations in these constructs using visualizations tracking scores across time. Below I use simulated responses across a few hundred data points to show how plotting the overall financial health metric－along with the constructs that make up financial health－can clarify why changes in financial health may be occurring. In this case, Use seems to track most closely with the overall metric, and Grow is exhibiting the steepest drop across time. Importantly, the bars indicate 95% confidence intervals, which in this case indicate that the observed shifts are not statistically significant.

Using Financial Health to Better Serve Customers

Aside from enabling us to track the financial health of our users, this research and scale development will help Tala more accurately segment users based on their overall financial health scores as well as their scores to each scale of financial health and happiness. For example, I will introduce and describe two hypothetical users from the same country with the same overall financial health score, but with different profiles across constructs: User A scores significantly higher on the Use scale compared to User B, while scoring significantly lower on the Protect scale relative to User B. Given their different scores on the Use vs. the Protect scales, each user would likely be categorized into different segments by a clustering algorithm.

Ultimately, the ability to segment users based on their financial health scores will give Tala the power to evolve our current products and to design future products that target the needs of different user segments. For example, given User A’s lower score on Protect, we infer that User A has reduced capacity to prepare for financial emergencies, and therefore an insurance product might be especially relevant to User A’s needs. In contrast, given User B’s lower score on Use, we might infer that User B has reduced capacity to safely store and to access their money for daily expenses, in which case a digital wallet product would be most appropriate to address User B’s needs. Without measuring financial health and the constructs that define financial health, we would not have the ability to personalize product offerings to users based on their financial needs.

User segmentation and personalized product offerings are only a few applications that become possible through the quantification and measurement of financial health. Moving forward, Tala is excited to continue embedding financial health in our business strategy and product development, while using specific metrics to track our progress towards our mission to increase financial access, choice, and control for our users.