# Lesson 6: Analysis Frameworks and Conceptual Models *Companion-podcast transcript • Sarah & Kiffer* *Office Hours episode to listen to after working through the lesson* --- **Sarah:** Here's a line from an interview we made up for this episode. An older man says, "My daughter calls every Sunday. It's so good to hear her voice. Then I sit with the phone in my lap for a long while." Two analysts code it. One tags it family contact as support. The other tags it contact that makes the absence louder. Which of them got it right? **Kiffer:** Maybe both of them, and so far neither one has made a claim. Those are two labels. Neither label says why he sits there holding the phone, or what that moment means across the rest of the interviews. Getting from a tidy label to a claim you can defend is the step a lot of analysis quietly skips. **Sarah:** And should two careful readers end up agreeing about a line like that? **Kiffer:** That depends on what you think a reading is for, and it's a question you and I don't fully agree on. **Sarah:** Welcome back to Office Hours. I'm Sarah. **Kiffer:** And I'm Kiffer. That step from labels to claims sits right at the centre of Lesson six, Analysis Frameworks and Conceptual Models, so let's start there. **Sarah:** Question one. When does a stack of codes turn into an interpretation? I ask because I think a lot of us believe we've made that move when we haven't. **Kiffer:** Let's test it with a made-up study. Imagine fourteen interviews about loneliness with grain farmers in Saskatchewan. After coding, four codes stand out: long days alone in the tractor cab, neighbours gone as farms got bigger and fewer, talk at the café in town that stays on weather and prices, and late-night phone calls with a sister. Sarah, write me the interpretation. **Sarah:** Okay. Farmers in this study described loneliness connected to long hours working alone, the loss of neighbours, and limits on what can be said at the café, and they coped through phone calls with family. **Kiffer:** Read it back and ask one question. Is there anything in that sentence that isn't one of the four codes with some grammar around it? **Sarah:** No, there isn't. It's the four codes in a row. I filed them neatly and called it a finding. That's the librarian from the lesson. **Kiffer:** And it's a very common version of that, because it sounds like a result. An interpretation says which codes matter most, how they connect, and why. So try again, and let yourself make a claim that could turn out to be wrong. **Sarah:** All right. In these transcripts, time alone in the cab comes up as an ordinary part of the work, and nobody calls it lonely. What people call lonely is having nowhere to talk about strain. As neighbours left, the café became the main place to see anyone, and the talk there stays on weather and prices. So the loneliness these farmers describe is about losing the places where worry could be said out loud. **Kiffer:** Now that's an interpretation. It ranks the codes, it links them, and it explains them. It also leaves the phone calls to account for. Where do they fit? **Sarah:** The calls to the sister would be the one place left where strain can be spoken. That would support the claim, if the transcripts show people saving the hard conversations for those calls. **Kiffer:** And that "if" is where the theoretical memo comes in. You put the claim at the top, list every transcript that supports it, and then list, just as carefully, every transcript that complicates it. **Sarah:** So here's a question for everyone listening. Describe a farmer whose interview would make me revise that claim. What would that farmer have to say? Take a few seconds, and pause the audio if you'd like more time. *(Pause)* **Kiffer:** Here are three that would do it. One would be a farmer who talks about strain every evening with a spouse or a close friend and still says he's lonely. Another would say the hours alone in the cab are the loneliest part of the year. A third might say the café is exactly where she talks about the hard things. Each one breaks a different link in the claim. **Sarah:** And then I'd have the three choices from the lesson. Revise the claim, narrow its scope, or treat the case as marking a boundary. **Kiffer:** Right, and the case itself often suggests which. Suppose the farmer who finds the cab lonely is twenty-six and took over the farm alone after a parent died. That might bound your claim to farmers who've run their operations for years, and it tells you to look for more farmers who are new to the work. It's also worth asking how much of what he feels in the cab is grief. What you can't do is leave him out of the memo because he's inconvenient. **Sarah:** Question two. When is a conceptual model a finding, and when is it a filing system with boxes drawn around it? **Kiffer:** The three standards from the lesson, grounded, transparent and revisable, do most of the work. Many first models fail earlier, in the way the categories or dimensions are set up. Here's one for everyone to diagnose. In a made-up study of why older adults in a mid-sized city don't come to a Sunday drop-in lunch program, an analyst proposes a taxonomy with five top-level categories: embarrassment about being seen as lonely, the cost of the lunch, never having heard of the program, no bus on Sundays, and structural barriers. Which standard for a taxonomy does this break, and how would you fix it? Take a few seconds. *(Pause)* **Sarah:** The categories sit at different levels of abstraction. Structural barriers is a big umbrella, and cost and the bus already belong under it. And no bus on Sundays is a single instance, one example of a transportation barrier. **Kiffer:** That's the problem. A fix would have two levels. One top-level kind is structural barriers, with cost and transportation under it. The other is personal or social barriers, with embarrassment under it, and probably more sub-kinds once you go back to the transcripts. There's also a decision hiding in there. Where does never having heard of the program go? **Sarah:** My first instinct is to put it under structural barriers and call it an information barrier. But you could also read it as the program's outreach failing, which is a different claim about whose problem it is. It might even deserve a place at the top level. **Kiffer:** And that's why placing a category is an interpretation. The memo that defends the taxonomy should say where you put it, why, and which transcripts pushed you there. A reader who disagrees can then disagree with something specific. **Sarah:** Typologies worry me more than taxonomies, honestly. A grid of sixteen cells looks so convincing once you start writing names into it. **Kiffer:** I'd run three checks on any typology. First, what is a case? If the same person shows up in two cells, your cases are episodes or accounts, and you should say so. Second, could you code each dimension without looking at the other one? If you call a trigger existential partly because the person talks about meaning, and you call a response reframing because the person talks about meaning, the link between them is built into your definitions. **Sarah:** So the pattern in that row would be something I put there while coding. The fix would be to define each dimension so it's coded from its own passages, and ideally have someone code one dimension with the other hidden. What's the third check? **Kiffer:** Count the cases in each cell. Twenty participants spread across sixteen cells is an average of one and a quarter people per cell. A pattern along a row made of one or two people per cell is a lead to follow with more sampling and close comparison of cases, which is where Lesson seven picks up. **Sarah:** What about empty cells? The lesson treats those as findings too. **Kiffer:** They're findings about this sample. Whether an empty cell tells you something about loneliness depends on whether your sampling gave that cell a fair chance to fill. With twenty people, I'd describe an empty cell as something we didn't see, and say what kind of participant might fill it. **Sarah:** Question three is about process models. The loneliness coping cycle from the lesson has a feedback loop, where the outcome of one episode changes how a person reads and responds to the next. But the loneliness interviews give us one transcript per person. Can a single interview show a cycle? **Kiffer:** That's worth working through before we answer it. Here's the question for everyone. If the feedback loop is real, what would a single transcript have to contain to support it? And what would make you doubt it? Take a few seconds. *(Pause)* **Sarah:** To support it, a person would need to describe at least two episodes and connect them. Something like, and I'm making this up, "The first time I tried a group it made things worse, so the next time I called my brother." To make me doubt it, I'd look for people who describe several episodes where nothing seems to carry over from one to the next. Or people whose way of understanding their loneliness changed for a reason outside the cycle altogether. **Kiffer:** Both are good answers, and the second kind matters a lot. If someone says a diagnosis, or a counsellor they were seeing for something else, changed how they understood their loneliness, the change came from outside the loop. The model needs room for that, or it needs to say plainly that it leaves that out. **Sarah:** There's still a problem, though. Even when someone links two episodes, they're telling me about the past. People tidy their history into a story when they tell it. **Kiffer:** They do, and that's a real limit. With one interview per person, the honest claim is about how participants describe their coping changing over time. That's still worth knowing. If you want to show the cycle as it unfolds, you need data collected over time, such as repeat interviews or diaries. That's data triangulation for a process claim. **Sarah:** Is there a classic process model that was built that way? **Kiffer:** Some of the earliest grounded theory came from Glaser and Strauss's work on dying in hospitals in the 1960s. Their book Awareness of Dying, published in 1965, was based on long observation of the care of dying patients in six California hospitals. Because they watched care unfold over time, they could treat dying as a process in its own right, which is the idea behind their later work on dying trajectories. **Sarah:** And there's a quantitative version of that kind of data here in Canada. The Canadian Longitudinal Study on Aging recruited more than fifty-one thousand people aged forty-five to eighty-five, and it collects data from them every three years, including a question about loneliness. **Kiffer:** Data like that can show whether loneliness comes and goes in the same people from one wave to the next, which bears on one part of the cycle. The waves are three years apart, which is coarse compared with a single episode. It can't reach the naming and interpretation stage, which is where interviews do their best work. Used together, each covers a gap in the other. **Sarah:** Question four. This one is a model I'd be pleased with, which is exactly why we should check it. It's a made-up study of twelve men who retired in the last two years, each interviewed once about loneliness. From the coding, I've drawn a concept map with four boxes: loss of the daily work routine, fewer contacts as work friendships fade, loneliness, and staying home more. **Kiffer:** And how do the arrows run between those four boxes? **Sarah:** Loss of routine leads to fewer contacts. Fewer contacts lead to loneliness. Loneliness leads to staying home more. And staying home feeds back into fewer contacts, which closes the loop. Every box is grounded, too. Loss of routine is coded in eleven of the twelve transcripts, fewer contacts in eleven, loneliness in nine and staying home in eight. So here's the question for everyone. Before you accept this map, what would you check? Take a few seconds. *(Pause)* **Kiffer:** I'd check the arrows, one at a time. The box counts show that each idea is in the data. Two codes turning up in the same transcript is co-occurrence, and an arrow claims more than that. It says one thing leads to the other, so it needs passages where someone connects the two. **Sarah:** So for each arrow, I go back and look for a man linking the two things in his own words. Something like, and this one is invented, "Once I stopped going in, I lost touch with the guys from the plant." That would count for the first arrow. **Kiffer:** Right. Lay it out as a small table, with the four arrows down the side and the twelve men across the top, and mark each man who describes that link. What does your made-up table show? **Sarah:** Eight of the twelve link losing the routine to losing contacts. Four link fewer contacts to loneliness. Loneliness and staying home turn up together in six transcripts, but only two of those men say the loneliness came first. The other four say they started staying home, and the loneliness followed. And nobody describes staying home cutting their contacts. I drew that arrow because it made the loop close. **Kiffer:** Then that arrow is a hypothesis drawn as a finding. The grounding standard covers every arrow as well as every box, so either take it out or keep it with a label that marks it as a hypothesis to look for. What would you do with the third arrow? **Sarah:** I'd make it two-way, and say in the memo that four men described staying home first and two described loneliness first. If staying home usually comes first, that's a different point at which a program might step in. **Kiffer:** There's one more trap, and it's the hardest one to see. Read the first two arrows as a chain: lost routine, then fewer contacts, then loneliness. How many of the twelve men describe that whole chain themselves? **Sarah:** Let me check the table. Only one of them does. The eight who link the lost routine to fewer contacts and the four who link fewer contacts to loneliness are almost entirely different men. Of the seven who describe only the first link, three say they aren't lonely, and four tie their loneliness to something else. **Kiffer:** So the map joins links reported by different people into a pathway that only one man actually described. It might still be true, but these data don't show it. Every box was grounded, the first two arrows were grounded, and the chain as a whole rests on a single case. **Sarah:** So the fix is to draw a small map for each man first, and then compare them. I might find two or three different pathways, which would be a more honest model than one composite. **Kiffer:** And the arrow table goes into the memo, so a reader can see what each line rests on. One caution from question three still applies. A man linking two things is giving you his account of the link, which is good evidence about how he understands his retirement and weaker evidence about causes. **Sarah:** Okay. Let's go back to the man with the phone in his lap, because this is where we disagree. **Kiffer:** Two analysts read the same line two ways. One says family contact as support. The other says contact that makes the absence louder. In investigator triangulation, I'd treat that disagreement as a signal to go back to the transcript and work out which reading the rest of the interview supports. **Sarah:** And I'd say the line supports both, and the analysis should keep both. He loves hearing her voice, and the call leaves him sitting alone with the phone. That's Laurel Richardson's point about the crystal. Different readers catch different faces of the same thing, and pushing them to converge can throw away exactly what's interesting. **Kiffer:** I'm sympathetic, but in applied public health someone usually has to act on a finding. If a health authority is deciding whether to fund regular phone contact for older adults, "both readings are true" is hard to use. I'd want the analysts to read the rest of his interview. Does he describe the calls as the high point of his week, or as a reminder of everything that's missing? Often the wider transcript tips the balance. **Sarah:** And sometimes it doesn't, and then the ambivalence is the finding. There's a real example of taking that seriously. A team studying loneliness among older people in Australian care homes used crystallization on purpose. They wrote sociological accounts of two cases, invited a novelist, Josephine Wilson, to write creative narratives from the same cases, and then compared what each form showed. **Kiffer:** That's a legitimate design, and it's open about what it's doing. My worry is the study that writes "triangulation" in its methods section and leaves it there, when two people from the same lab coded a few transcripts and settled their disagreements by discussion. That can be a fair start on investigator triangulation, but agreement between two analysts trained together is weak evidence on its own. And a study can also lean on crystallization to avoid ever deciding between readings. **Sarah:** I agree with that. Crystallization done well takes more work. You still have to show each face with data, and say where the faces conflict. **Kiffer:** So here's where I land. For most applied public health studies, I'd use triangulation and aim to say what the data support. When two readings both hold up in the data, I'd report both and say what each would mean for practice. **Sarah:** I'd go further than you toward keeping the facets, but we agree on the practical rule. Name the check you actually ran, who did it and what it changed. And when readings diverge, report whether you resolved them with more data or kept both, and why. **Kiffer:** Let's pull it together with three things to take away. **Sarah:** First, a list of codes with grammar around it is still a list. An interpretation ranks the codes, connects them and explains them, and it's specific enough to be wrong. **Kiffer:** Second, ground the arrows as well as the boxes. A code that appears in many transcripts shows that an idea is there. An arrow needs passages where people connect two things, and a chain needs people who describe the whole chain. **Sarah:** And third, match the claim to the data. One interview per person supports a claim about how people describe change, and watching a process as it happens needs data collected over time. Where the data are thin, like a typology cell with one person in it, say so and treat it as a lead. **Kiffer:** If you'd like more practice, go back to the grain farmers from question one. Write a short theoretical memo for the second version of the claim, describe the transcript that would make you revise it, and say which of the three responses you'd choose. **Sarah:** Next time, it's Lesson seven, Comparing Variables and Grounded Theory, where constant comparison and matrix displays give you tools for comparing cases systematically. **Kiffer:** Take care, everyone. **Sarah:** See you in Lesson seven.