"We ran sentiment analysis on the open-ended comments" and "we ran thematic analysis on the open-ended comments" get used almost interchangeably in casual conversation about survey and feedback data, as if they were two names for the same underlying work. They're not. They answer genuinely different questions, produce genuinely different kinds of results, and - this is the part that causes the most confusion in practice - a strong result from one doesn't tell you much at all about the other.
Table of Contents¶
- Two Different Questions
- Why People Conflate Them
- What Each One Actually Misses on Its Own
- Why the Combination Is Usually the Real Answer
- FAQ
Two Different Questions¶
Sentiment analysis answers: how did people feel? Positive, negative, neutral, or some more granular scale in between - it's a measure of emotional tone, and its natural output is a distribution: 62% positive, 24% neutral, 14% negative. It doesn't, on its own, tell you what any of those people were actually talking about.
Thematic analysis answers a different question: what were people actually saying? It sorts responses by topic, subject, or underlying idea - pricing, onboarding, a specific feature, a particular kind of interaction - regardless of whether each theme was raised positively or negatively. Its natural output is also a distribution, but a distribution of subjects, not tones: 34% mentioned pricing, 28% mentioned onboarding, and so on.
A response can be strongly negative and about almost anything; a response can be about pricing and be either strongly positive or strongly negative. The two dimensions are independent of each other, which is precisely why treating one as a stand-in for the other loses real information.
Why People Conflate Them¶
Part of the confusion is genuinely reasonable: in casual conversation, "how did people feel about X" quietly bundles both questions into one sentence, since a full answer to "how did people feel" naturally wants to specify both the tone and the topic at once - "people felt frustrated about pricing" is really two findings stitched together, one from each kind of analysis. It's easy to run just one of the two, get a clean-looking result, and describe it using language that implies the other was covered too, without anyone involved intending to overstate anything.
There's also a practical reason sentiment gets treated as the easier, more complete-feeling answer: a single positive/negative/neutral distribution is simpler to compute, simpler to chart, and simpler to present in one line of a dashboard than a full thematic breakdown, which makes it tempting to lean on as if it were answering the fuller question on its own.
What Each One Actually Misses on Its Own¶
Sentiment analysis alone tells you that something is wrong (or right) in aggregate, without telling you what to actually do about it - a report that says "sentiment dropped from 71% to 62% positive this quarter" raises the obvious next question (about what, specifically?) without answering it, and a reader is left needing a second analysis just to know where to look.
Thematic analysis alone tells you what people are talking about without telling you whether that's a good or bad thing - "34% of responses mentioned the mobile app" is topically informative and emotionally silent; the same topic could be dominated by praise or by complaints, and the theme breakdown alone genuinely can't distinguish the two. A theme list without any sentiment attached to it is a table of contents for a report that hasn't actually said anything yet about whether those topics are problems or strengths.
Why the Combination Is Usually the Real Answer¶
Most of the findings people actually want from open-ended data live at the intersection of the two - not "people are unhappy" (sentiment alone) and not "people talk about pricing" (theme alone), but "people are unhappy specifically about pricing, while being largely positive about everything else," which requires both dimensions crossed against each other to state. This is why a mature approach to open-ended analysis usually treats sentiment and theme as two separate classifications run on the same data and then combined, rather than trying to force one classification to do both jobs simultaneously - a genuinely different design decision than, say, running a single classification with categories like "positive pricing" and "negative pricing" bundled together, which tends to produce a sprawling, harder-to-maintain category list compared to two clean, separately-defined classifications crossed together after the fact.
Braun and Clarke's influential thematic analysis framework, one of the most widely used structured approaches to qualitative theme identification, is explicitly about identifying patterns of meaning across data - not about scoring emotional tone, which is a related but distinct research tradition with its own separate methods and history. Treating the two as genuinely separate lenses, each answering part of the picture, tends to produce a clearer, more specific finding than expecting either one to carry the whole analysis alone.
FAQ¶
If I can only run one, which is more useful?
Thematic analysis is usually more actionable on its own, since it points toward a specific subject to investigate further, while sentiment alone often just confirms that something needs attention without saying what. That said, running both and crossing them is a meaningfully stronger result than either alone whenever it's feasible.
Is sentiment analysis a subset of thematic analysis, or the other way around?
Neither is a subset of the other - they're parallel, independent dimensions of the same data. A response's sentiment doesn't determine its theme, and its theme doesn't determine its sentiment, which is exactly why combining them (rather than nesting one inside the other) produces the fullest picture.
Does "mixed methods" mean the same thing as combining sentiment and theme?
No - "mixed methods" is a broader research term referring to combining qualitative and quantitative approaches generally. Combining sentiment and thematic classification is a narrower, specific technique within qualitative analysis, not the same concept.
Can a single response have one sentiment but multiple themes?
Yes, and this is common - a lengthy open-ended response might touch several distinct topics while carrying one overall tone, or occasionally a genuinely mixed tone as well. Handling that combination well is exactly why sentiment and theme are usually best run as two separate, crossable classifications rather than one combined pass.
For a practical guide to setting up each of these as a classification, see Building a Sentiment Classification Without a Dedicated Sentiment Tool and Writing a Good Classification Goal.