Connecting Qualitative Personas to Population Data

Personas you can count

The starting point: five dimensions of digital life

Digital Personas was built through two rounds of qualitative research across Kenya, Nigeria, and Senegal — over 250 interviews with women, and around 100 more with the husbands, children, and peers who shape their digital lives. The research was built on Pathways, a population-representative segmentation of women in these countries: we used its segments to screen and recruit participants, and later to analyse the data. We synthesised the findings into personas describing distinct modes of digital engagement, from women who rely entirely on others to use a phone, to women who navigate complex digital systems independently.

Each persona is built on the same five dimensions: relevance (why digital matters in her life), skills (what she can do and how she learned it), affordability (what devices, data, and repair cost her), safety (what risks she perceives and how she manages them), and norms (what she is permitted, expected, and scrutinised for).

Fig 1: Five interconnected dimensions of women’s digital lives

So far, this is what good design research produces. The interesting part is what came next.

The problem with the obvious three variables

We wanted to market-size the personas — to say not just who these women are, but how many of them exist, and where, which meant anchoring them in population-representative data. The Demographic and Health Survey (DHS) was the natural choice: nationally representative, comparable across our countries, and trusted by the institutions whose decisions we want to shape.

The DHS offers only three variables on women’s digital lives: mobile ownership (with smartphone ownership added in recent rounds), internet use, and mobile finance. We could have typed personas on those three, but our fieldwork told us these are outcome variables. They record whether a woman has a phone — not whether she can charge it, pay for data, repair it, or use it without scrutiny. Access does not equal use, and use does not equal agency. For example, among women in urban Northern Nigeria who own smartphones, the classic indicator of digital inclusion, 25% cannot read a full sentence. Two women can tick the same boxes and live entirely different digital lives. Whatever produces those different lives sits upstream of these indicators.

Going upstream

To find those upstream drivers rigorously, we ran an association analysis across the DHS: measuring (with Cramér’s V and Spearman’s rho) which variables in the survey move most strongly with the three digital indicators. The analysis answers a simple question: when a woman uses the internet, what else tends to be true about her life?

The analysis surfaced two kinds of variables:

1. Variables that move with the digital indicators but don’t drive them. Having a bank account, for instance, moves with phone ownership, but a bank account doesn’t necessarily produce digital capability.
2. Variables that move with the indicators and sit upstream of them. Education is the clearest case: our fieldwork showed repeatedly how education shapes digital confidence, literacy, and independence.

Fig 2: Association Heatmap in Senegal showing top variables associated with digital outcome variables (owns mobile, uses mobile finance, owns smartphone, uses internet)

This is something we want to emphasize — association analysis alone cannot tell these two types apart. It measures correlation, and correlation treats a bank account and education identically. What let us separate drivers from co-travelers was two years of fieldwork. For example, we had traced how household income shapes device access, data purchase, and repair decisions — so when the wealth quintile showed high association, we knew the mechanism behind the number. The qualitative research is what made the quantitative analysis interpretable. That is the methodological point of this post.

Fig 3: Connecting qualitative insights with population-level data

The drivers also varied by geography, in ways that matched what we heard in the field. Partner’s occupation, for example, was far more tightly associated with women’s digital indicators in Senegal than in Kenya. A single global model would have flattened the differences our research existed to surface, so we didn’t build one.

Typing, not clustering

With the digital indicators and their upstream drivers in hand, we built a rule-based typing tool (a set of human-readable rules that assigns every eligible respondent in the DHS to a persona) for each geography, which too were anchored in our qualitative data.

If you’ve worked with quantitative segmentation, you will notice what we didn’t do. We didn’t run a clustering algorithm on the survey and interpret the emerging clusters. We also didn’t train a predictive classifier. Instead we wrote the rules ourselves, through a qualitative process of mapping each persona’s defining attributes to the DHS variables the association analysis had validated.

Fig 4: A section of the rural Senegal typing tool. Please note that this is not the entire rule-set

This was a deliberate choice, and has both pros and cons. Our typing is deterministic, as the same woman always types the same way, but it doesn’t claim the statistical elegance of a fitted model. In exchange, every rule is legible. A sceptical stakeholder can open the typing tool and see exactly why a respondent is assigned one persona and not another, and argue with us about it.

Our typing is deterministic, so the same woman always types the same way. However this approach does not yield a measure and a probability of fit that a classifier would. A woman near a rule’s boundary is assigned as firmly as one in the middle of a persona, and the boundaries themselves are researcher judgement, not optimised thresholds. That judgement, though, is the point as the thresholds come from what we learned in the field, which is where the qualitative and quantitative actually meet, rather than one being tacked onto the other. And it makes every rule legible. A sceptical stakeholder can open the typing tool and see exactly why a respondent is assigned one persona and not another.

What this unlocks

Locating personas inside the DHS changed what they can do:

Prevalence. We can estimate how many women fit each persona — nationally, and down to state and district levels. (We have an upcoming post about this estimation process.)

Quantitative depth on qualitative personas. Every persona now carries attributes from the DHS: literacy rates, participation in household decision-making, exposure to domestic violence, wealth distribution. The persona dashboards on designwithdp.org show this layering — a narrative grounded in fieldwork, sitting alongside population statistics for the women that narrative represents.

Mapping to other segmentations. Because personas live in the DHS, we can map them onto the Pathways population segments. The mapping is a matrix, not a one-to-one lookup: a single Pathways segment contains multiple modes of digital engagement, which is itself a finding. Pathways tells you who you’re designing for; the personas tell you how those women engage with digital.

And because the typing runs on the DHS, it can run again whenever the DHS updates. We’ve already re-typed personas on the 2024 Nigeria survey and watched market sizes shift since 2018.

What we’re not claiming

The typing rules are deterministic, but the mapping behind them is interpretive, built on researcher judgement and insight. The personas are modes of engagement, not boxes women live in; a woman’s mode can shift as her circumstances do. And a persona in Lagos is not directly comparable to the same persona label in rural Kenya, because the conditions that shape digital life differ.

We think of this less as a finished method and more as a practice we’re building: qualitative research that produces the understanding, quantitative analysis that gives it reach, and a transparent, arguable bridge between the two. Design research that can survive the question “how many of them are there?”, without giving up the answer to “who are they, really?”

Digital Personas is developed by Quicksand and Stby, and funded by the Gates Foundation. We’re grateful to Tracy Johnson at the Gates Foundation, who we collaborated with closely in shaping this methodology. Explore the personas at designwithdp.org. Write to us at digital-personas@quicksand.co.in.