From Regression to Reflection: A Mixed-Methods Journey

In the words of Brené Brown, “The clean lines of quantitative research appealed to me, but I fell in love with the richness and depth of qualitative research.” Coincidentally, Professor Brown has also extensively researched the concept of shame—something I strongly felt when my quantitative model aimed at predicting inflation’s effect on different income groups didn’t explain as much as I had hoped. I tested for multicollinearity and all the diagnostics seemed fine. The limitation, however, was clear: I was working with just 12 months of data leading up to February 2025.

But in moments like these, where policy windows are narrow and urgency is high, we don’t always have the luxury of longitudinal design. What if the problem wasn’t the model but the frame? What if the real question was: How can researchers make meaningful contributions with limited, short-term data—especially when the last year alone has seen profound economic and social shifts? 

So, I reframed my research project. Inflation became a case study—not the main event. I turned the spotlight onto the method itself: Can quantitative research help us say something useful in the short term? Can qualitative insights fill the gaps? Does the sequence in which we interpret these (qualitative>quantitative or vice versa) impact our conclusions?

Quantitative Analysis

I used multiple and quantile regressions to see how inflation was affecting different income groups. On the surface, incomes looked like they were rising—probably due to wage adjustments. But real earnings weren’t keeping pace. People with higher debt-to-income ratios also reported higher incomes, suggesting many were leaning on credit to stay afloat. Spending more often correlated with earning less, signaling financial stress. Even when unemployment rose, income figures held steady—likely because job losses hit lower-wage sectors hardest.

Qualitative Analysis

I analyzed reports from the OECD, IMF, and World Bank, which revealed a clear shift: governments are leaning more on fiscal tools, and inflation strategies vary by country. More importantly, they showed how much economic outcomes depend on public confidence, especially among vulnerable groups. Inequality isn’t a background issue anymore—it’s central to how stable or fragile a recovery really is.

Sequencing: What Comes First?

Now, most of us have a preferred sequence. Some lean on numbers before narratives; others lead with voices and stories. So, I handed the job over to a large language model (LLM), prompting it to interpret my combined findings in both sequences—quant→qual and qual→quant. I repeated each combination a couple of times and compared how the insights shifted.

I didn’t label what was quant or qual in my prompts and stripped away any sequencing cues—so the model wasn’t led by structure or wording.

The LLM consistently generated stronger insights when starting with quant. For example:

This version integrated hard data with socio-political outcomes, weaving them together rather than placing them in juxtaposition—a contrast to the separation that the qual→quant sequence tended to produce:

Here, key links were treated as afterthoughts rather than explored.

Wait… Is That It?

Thinking I was done with my research at this point, I looked back to where it all began, remembering how often I’ve told researchers during the academic training sessions I conduct as a part of my job: “Don’t be disheartened if your data doesn’t yield the results you hoped for.” It’s advice I’ve genuinely believed, but when I found myself grappling with results that didn’t explain everything, I didn’t want advice. I just wanted to be listened to.

This realization shifted something. It reminded me that research, especially in uncertain times, isn’t just about patterns; it’s about listening. And that’s where surveys came into the picture. I rolled out two surveys: a closed-ended one for structure and an open-ended one that allowed people to share how inflation was showing up in their daily lives.

Survey Responses

Both surveys showed what the models couldn’t. Middle-income earners bear the brunt too, even though their unique struggles are often masked when they are clubbed with low-income earners in some studies; 40% said expenses rose 10–20%, and 80% said salaries weren’t keeping up. In their own words, they talked about cutting back, picking up side gigs, delaying big purchases—and feeling a creeping loss of financial control. Lower-income participants described stark difficulties as well: skipping meals, postponing medical care, and relying on credit to survive.

Summoning the LLM Again

The two-part rollout of the surveys was intentional: Just like with the macro-level analysis, I ran a similar sequencing experiment here too—instructing an LLM to interpret the findings a few times each in different orders: first closed-ended then open-ended, and vice versa. Once again, I compared the outputs. A similar pattern emerged. Starting with closed-ended data followed by open-ended responses yielded more layered insights than the reverse order. For instance:

The deeper insights into the “why” behind policy preferences conveyed by “viewing inflation as a crisis requiring immediate relief” and “perceiving inflation as a systemic issue that needs structural fixes,” was missing in outputs for the reverse order:

The Answer or the Ask

In a nod to Professor Brown’s Daring Greatly, I finally put this research out into the world, as a project by a researcher for other researchers navigating complexity and change. To ensure complete transparency, I’ve shared the full LLM chat transcripts in the manuscript (and I do so here too, as a parting gift). Understanding inequality isn’t just about measuring it—it’s about asking who cares, why they care, and what they’re willing to change. And in doing so, I realized that how we interpret data—especially in volatile, unequal times—depends as much on sequence and structure as on substance. That doesn’t make the search for answers easier. But daring greatly isn’t about having all the answers. It’s about being willing to ask better questions—and letting the data, and the audience, speak back.