# Lesson 7: Comparing Variables and Grounded Theory *Companion-podcast transcript • Sarah & Kiffer* *Office Hours episode to listen to after working through the lesson* --- **Sarah:** Here's a sentence I made up, though I suspect you've read one like it. "Using a grounded theory approach, we interviewed eighteen older adults, and five themes emerged from the data once saturation was reached." Kiffer, what do you hear? **Kiffer:** I hear three claims folded into one sentence. Grounded theory is supposed to end in an explanation organized around a core category, and this one ends in five themes. The word emerged hides the person who built those themes. And the claim that saturation was reached is something a reader can't check unless the paper says what stopped turning up, and when. **Sarah:** And yet every piece of that sentence names something a careful analyst could actually do. **Kiffer:** Which is why it's worth taking apart. Each of those claims points to a method, and a reader needs to know whether the method was actually used. **Sarah:** Welcome back to Office Hours. I'm Sarah. **Kiffer:** And I'm Kiffer. That made-up sentence sits right at the centre of Lesson seven, Comparing Variables and Grounded Theory, so let's start there. **Sarah:** Question one. What makes an analysis grounded theory, and what should you call it when you can only do part of it? **Kiffer:** The clearest test is the end product. Grounded theory ends in a substantive theory, an account of why a phenomenon takes the shape it does, built around one core category, with the other categories related to it in ways you can state. A list of themes with quotations describes what's in the data. That's a perfectly good finding, and it has its own name. **Sarah:** But the end product can't be the whole test. I could write a core category on top of a list of themes and call it a theory. **Kiffer:** True, so the second test is the procedure. Constant comparison has to run through every stage, at all four levels from the lesson: within an extract, within a case, across cases, and from the category back to the data. Theoretical sampling has to decide what gets read or collected next. And memos have to carry the categories as they develop. Without those commitments, a core category is a label sitting on top of a description. **Sarah:** And on a fixed dataset, like the course's loneliness interviews, theoretical sampling becomes a decision about which transcript to read next. You write down the order and the reason for each choice, so a reader can see it in the methods section. Which brings me to the word emerged. Where does it even come from? **Kiffer:** It has a history. Barney Glaser used emergence to mean categories arising from the data without being forced into a structure decided in advance. When Anselm Strauss and Juliet Corbin published their coding procedures in 1990, Glaser's reply in 1992 framed the whole dispute as emergence versus forcing. So the word carries a position with it. **Sarah:** And Glaser and Strauss came to grounded theory from quite different places, if I remember right. **Kiffer:** They did. Glaser trained at Columbia under Paul Lazarsfeld, the survey methodologist, and Robert Merton. Strauss trained at the University of Chicago, in the symbolic interactionist tradition. Kathy Charmaz, who studied with both of them, later argued that analysts are part of the world they study, and that they construct their theories through their interactions with participants. So a paper that claims Charmaz's constructivist approach and then says its themes emerged is describing its own analysis in two ways that sit uneasily together. **Sarah:** It's worth saying why public health researchers want the full version at all. Grounded theory is at its best on process questions, like how people come to seek help for their mental health, or how stigma plays out in someone's life over time. Those are questions about how something unfolds, and a count of who said what can't answer them. **Kiffer:** And that's where the label matters most. A paper on help-seeking that calls itself grounded theory is promising its readers an explanation of the process, and they're entitled to look for one. **Sarah:** Let's test all this on a full methods paragraph. Again, this one is invented. Eighteen adults over sixty-five were recruited from two seniors' centres in one city and interviewed once each. After the first three interviews, two analysts built a codebook, then applied it to all eighteen transcripts. Five themes emerged, and saturation was reached. The authors call it a grounded theory study. If you were reviewing this paper, what would you ask them? Take a few seconds, and pause the audio if you'd like more time. *(Pause)* **Kiffer:** I'd start with the procedure, because it already has a name. Building a codebook from the first few transcripts and applying it to all the rest is close to the framework method's sequence from the first half of this lesson: develop a working analytic framework, then apply it to every remaining transcript. That's a sound way to work. It's also a different method. **Sarah:** So the first question for the authors is which method they actually used. My second would be about sampling. All eighteen people came from the same two centres, and nothing in the paragraph says the early analysis changed who was recruited next. **Kiffer:** Right, so I'd ask whether recruitment changed as the categories developed. If it didn't, I'd ask in what order the transcripts were read, and what analytic reason drove that order. Third, where is the explanation? Five themes tell me what's there. I'd want a core category and the relationships around it. **Sarah:** And saturation. I'd want to know which categories stopped gaining new properties, and at which interview. Plus one more question. Which cases didn't fit, and what happened to them? **Kiffer:** Good, that's five questions. Now suppose the honest answers are that they built a codebook, recruited everyone in advance and wrote up themes. Then the fix is a relabel: a codebook-based thematic analysis, or the framework method if they charted a matrix. If they did use constant comparison and memos toward a core category but couldn't sample theoretically, the lesson's label fits: a grounded-theory-informed analysis. In my view, an accurate label makes a study more credible, because it tells readers exactly what to expect. **Sarah:** Question two. What does open coding look like when it's done well, and how do you keep the categories you already know from taking over? I'd like to try this one myself. **Kiffer:** Okay. Here's a short excerpt I've invented for this. Picture a man in his fifties who drives a delivery route. He says, "I've got the guys at the depot. We talk hockey, it's fine. My wife passed two years ago, and nobody there knows. I don't bring it up. In the truck I keep the radio on all day. Some nights I call my sister and barely say anything. I just want to hear her kids in the background." Before Sarah starts, try the middle of that, "nobody there knows, I don't bring it up." Give it a code in the form of an action word, like naming or avoiding. Take a few seconds. *(Pause)* **Sarah:** Okay. "I've got the guys at the depot" is having workplace company. "We talk hockey, it's fine" is keeping talk on safe topics. Then "nobody there knows" and "I don't bring it up." I'd code that as recognition failure inside available presence. He's around people every day who can't recognize his loss. It's the phenomenon from the worked analysis in the lesson. **Kiffer:** Look at that last code again and tell me two things about it. What form is it in, and where did it come from? **Sarah:** It's a noun phrase, and it came from the lesson. Another analyst built that category at the axial stage, from different transcripts, and I attached it to one line of a new one. That's a small version of what Glaser called forcing. **Kiffer:** It is. And there's a cost to that label beyond the form. Read his words once more and see what it covered up. **Sarah:** "I don't bring it up." He's choosing. The category says the people around him can't give him recognition, and in his own words he hasn't given them the chance. So I'd give that part two codes. One is noting that nobody at work knows about his wife's death, and the other is choosing not to disclose the loss. **Kiffer:** That's the level open coding works at, what he's doing in the talk. Keep going with the last three sentences. **Sarah:** The radio I'd code as filling the silence in the truck. The calls as calling without talking. And the kids in the background as listening for the sound of family life. That one's interesting. He wants company with almost no conversation in it. **Kiffer:** And now you have something real to compare. At the category-to-data level of constant comparison, you can ask whether the core category from the worked analysis holds for him. Withholding the loss might be a way recognition fails that the category doesn't yet name. The call to his sister is harder to read. The excerpt doesn't say whether she knew his wife. If she did, a quiet call might be recognition that needs very few words. That's a question for a memo, and you only reach it because your first pass coded what he did. **Sarah:** So where would the paradigm model come in? **Kiffer:** At the second stage, once you have open codes from several transcripts. Try placing what he said in Strauss and Corbin's template. **Sarah:** I'd name the phenomenon first. Call it carrying a loss alone among people he sees every day. His wife's death is a causal condition. The depot is context. Keeping the loss to himself, the radio and the calls are his strategies. The consequence might be the quiet company he gets from those calls, though I'd want more of the interview before I wrote that down. **Kiffer:** That's a reasonable first placement. Putting his silence among his strategies fits your recoding, and your caution about consequences is right, because the excerpt barely shows any. Charmaz's focused coding works differently. You'd pick your most productive open codes, say choosing not to disclose, and re-read the other transcripts with them. Both are axial work, and both come after a first pass like yours. **Sarah:** Question three. When does a count in a matrix actually mean something? **Kiffer:** Some history first. In 1958, Howard Becker described the informal counting good fieldworkers do as quasi-statistics, a name he credited to Paul Lazarsfeld and Allen Barton at Columbia. So Lazarsfeld was Glaser's own teacher, and his survey tradition was one of the things Glaser and Strauss were arguing with. He also helped give qualitative researchers a name for careful counting. **Sarah:** And the rules from the lesson follow from that idea. Count when commonness matters to your claim, and report the denominator so a reader can check it. Give me a case where a count can still mislead you after you've done both. **Kiffer:** Here's a made-up study. Ten adult children caring for a parent with dementia, interviewed once each. Nine of the ten have brothers or sisters, and they mention them in passing almost every time they describe a week of care. Three codes. Talking about siblings appears in nine of the ten transcripts. Keeping the parent's old routines going, like making their tea the way they always took it, appears in four. Going unthanked, the sense that nobody notices the work, also appears in four. **Sarah:** Siblings is in nine of ten mostly because nine of the ten have siblings. That count tells you more about these families than about caregiving. And the tenth person's empty cell is a genuine absence, because they have no siblings. An empty cell for someone with a brother who simply never came up would mean something different, and the memo should say which kind of empty it is. **Kiffer:** Good. Now add the optional step from the lesson's matrix procedure. For every pair of codes, count the participants who have both. All four people who talk about routines also talk about siblings, so that pair is four. All four who describe going unthanked also mention siblings, so that's four as well. Routines and going unthanked share three people. **Sarah:** So siblings looks like the hub. It's linked to each of the other codes more often than they're linked to each other. **Kiffer:** That's what the raw counts suggest. So here's a question for everyone. If you could go back and read only one pair closely, which pair would you choose, and why? Take a few seconds. *(Pause)* **Sarah:** I'd ask how often each pair would overlap anyway. Siblings is in nine of ten transcripts, so almost any code will overlap with it. Among the four people who talk about routines, all four mention siblings. Among the six who don't, five mention siblings. That's one hundred percent compared with about eighty-three percent, barely a difference. **Kiffer:** Now do the same check for routines and going unthanked. **Sarah:** Among the four who talk about routines, three describe going unthanked. That's seventy-five percent. Among the other six, only one does, about seventeen percent. So the pair with the smallest raw count shows the biggest difference, and it's the one I'd read first. **Kiffer:** That's the trap. A code that turns up in almost every transcript will co-occur with almost everything, so its pair counts look impressive and tell you very little. And even the routines pair is only a pointer to where to read. Two codes in the same interview may never touch each other in the talk. You'd go to those three transcripts and see whether keeping the routine is part of the work nobody thanks them for, or whether the two come up in unrelated parts of the conversation. **Sarah:** And if they are connected, how would I report it? **Kiffer:** With the denominators and the context. Something like, three of the four participants who described keeping a parent's routines also described going unthanked, compared with one of the six who didn't describe routines. Then the extracts that show the connection. With ten people, one participant moves those percentages a long way, so it's a pattern worth a memo and worth testing in more cases. That's quasi-statistics in Becker's sense, a check on over-claiming. **Sarah:** Question four, and this is one where Kiffer and I don't fully agree. When a case doesn't fit your category, how do you tell whether you've sharpened the category or just stretched it? **Kiffer:** Let's use the case from the lesson. Frank is eighty-two and lives in long-term care. He describes his loneliness with a calm that, at first, doesn't fit a core category about the work of converting presence into recognition. The worked analysis reads his calm as that work completed, and adds a trajectory to the category, running from acute failure through chronic struggle to either settled adaptation or being stuck. **Sarah:** And I find that too convenient. Once the category covers people in acute distress, people struggling for years, people who've settled and people who are stuck, it can take in almost anyone. Negative-case analysis is supposed to protect us from confirmation bias, and the moment we reinterpret the one case that resists is exactly where confirmation bias would show up. At that point I'm not sure what it would take for the category to be wrong. **Kiffer:** That's a fair worry, and I'd still defend the move. Constant comparison is supposed to make a category denser, so each category carries a trail of instances it explains and instances it had to be refined to handle. This refinement adds a claim about order in time, and it names two uncoded transcripts to test it on. That's a category becoming more specific. **Sarah:** Specific in a way that can fail, though? If someone's still struggling after twenty years, they're stuck. If someone's calm after one year, they settled early. What case would ever count as evidence that the category is wrong? **Kiffer:** That's the right question, so let's give it to everyone. Taking the refined category as it stands, what kind of case would genuinely challenge it? Take a few seconds. *(Pause)* **Kiffer:** I can think of two. One is a person who is calm about being alone and describes no recognition work at all, someone who chose solitude as a practice. The lesson's own example of theoretical sampling, the long-distance hiker, is that kind of case. If calm can come from somewhere else, Frank's calm can't automatically be read as finished work. The second is a person whose loneliness eased through more contact alone, with no change in who recognized them. That would challenge the claim that strategies aimed at recognition do more work than strategies aimed at presence. **Sarah:** So the category can fail. I'll grant that. I'd still report Frank as a possible separate type in the write-up, something like calm acceptance, and let the next cases decide. Folding him in on the strength of one transcript feels early to me. **Kiffer:** I'd fold him in, provisionally, because the evidence for completed work is in his own transcript: morning coffee with Mavis in the lounge, and a student volunteer who comes on Saturday mornings. I read those as small, regular presences he has turned into recognition. If his transcript described none of that, I'd agree with you. **Sarah:** Then here's where we agree. Before you reinterpret a case that doesn't fit, write down in a memo what would have counted as a real exception. If you refine the category, point to the evidence in that transcript, and name the next cases that could break it. And in the write-up, say plainly that the case was reinterpreted, so a reader can disagree with you. **Kiffer:** Agreed. A negative case should leave a visible mark, either on the category or in the findings as an exception. The outcome to avoid is a case that gets quietly set aside. **Sarah:** Let's pull it together with three things to take away. First, name the method you actually used. A codebook applied to every transcript and written up as themes is a descriptive analysis. Grounded theory techniques on a fixed dataset make a grounded-theory-informed analysis. And any claim of saturation needs to say what stopped appearing, and when. **Kiffer:** Second, code what's in the line before you bring in categories you already know. Action words keep you close to what the participant is doing, and the categories you've read about get their test later, through constant comparison. **Sarah:** And third, a count in a matrix tells you where to look. Report the denominator, and before you call two codes connected, check how often one turns up among the people who have the other compared with the people who don't, because a code that's nearly everywhere will pair with nearly everything. **Kiffer:** If you'd like more practice, take the delivery driver's excerpt from this episode, code it yourself line by line, and then write a short memo on whether the core category from the lesson fits him, and what new property it might need if it doesn't. **Sarah:** Next time, it's Lesson eight, Content Analysis, where codes are treated as variables and the counting we did today becomes a method of its own. **Kiffer:** Take care, everyone. **Sarah:** See you in Lesson eight.