Academy / Foundations / Lesson 3
Additional course links
You leave with: A linked view of scores and open answers for one question, a joint display with its bases, and a finding sentence with its limit and next step.
Where this fits: Lesson 3 names mixed-method context, a score beside the person’s own words, as the kind most often lost. This page answers the second half of the training team’s question: what stopped the others? You bring back a linked view, a joint display and one finding sentence for your plan.
The funder’s question has two halves. Which completers used the skill at work after 30 days? is a count: of 40 completers, 25 answered the follow-up, 15 said Yes, 10 said No, and 15 are unknown. What stopped the others? lives in the open answer that follows: “What helped, or what got in the way?”
On the usual path the Yes/No column goes into a spreadsheet chart and the open answers into a separate tab. Paste both into ChatGPT and it cannot tell which reason belongs to whom; with hundreds of open answers beside the numbers, it summarizes the loudest. The reasons of the 10 who said No blur into those of the 15 who said Yes.
How do you connect quantitative and qualitative survey data?
In short: Keep each score and each open answer on the same ID, wave and question, then read them together against one clear question. A joint display puts the number, the words and your interpretation side by side. The words suggest explanations; they do not prove why a number moved.
When both answers come from the same form under the same ID, nothing needs joining. When they arrive from different tools, restore the link without guessing.
Why start with the question?
In short: The question decides the split. “What stopped the 10?” calls for comparing reasons by outcome; “How did learners experience the course?” calls for reading all the answers together.
The training team’s question supports a comparison between the 15 who used the skill and the 10 who did not. The 15 who did not reply are not a third group with reasons: there are no words and no answer, so they stay “unknown” in every table. Write the grouping rule before reading the comments. If a new comparison occurs to you while reading, keep it and label it exploratory. Fetters, Curry and Creswell’s integration framework describes other routes; the split is one option, not a requirement.
How do you link the right records?
In short: Link on the ID, not on names, and keep the wave, the question wording and who said it. The unit is one person per wave.
Maria’s open answer at 30 days is her own account. Her mentor’s week 8 note is someone else’s account of her. Both belong on ID 0417, labelled by who wrote them, never merged into one comment field. The same goes for waves: an exit comment explains the exit score, not the 30-day answer.
Two spreadsheet mistakes are common. Repeating a score once per comment gives talkative people more weight; count at the person level. And forcing uncertain matches to empty the unmatched pile hides who is missing; report the unmatched count instead. Anonymous feedback can still be read at the theme level without re-identifying anyone.
What does the joint display look like?
In short: First the linked view, one row per person with the answer and their own words. Then the joint display, themes by outcome group, with the base for each group written on the table.
A few rows from the training team’s linked view (fictional IDs and words):
| ID | Used the skill? | Their own words at 30 days | Other evidence on the ID |
|---|---|---|---|
| 0417 | Yes | “I used it my first week. Leading the mock interview made the difference.” | Mentor note, week 8: led a mock interview |
| 0452 | No | “My manager hasn’t given me work where it fits yet.” | Exit confidence 4/5 |
| 0388 | No | “I wasn’t sure I’d get it right in front of a real client.” | Exit confidence 3/5 |
| 0501 | Yes | “Tried it once. It felt rushed.” | Exit confidence 3/5 |
| 0430 | No reply | None | Counted as unknown, never No |
From the coded answers the team builds the joint display. The layout matters more than any single number in it: each group’s base is on the table, and the last column says what to check, not what the team concludes.
| Reason (theme) | Used it · base 15 | Did not · base 10 | What to check next |
|---|---|---|---|
| No time to practise | count of 15 | count of 10 | Shift patterns and workload; whether a practice slot after the course would help |
| No chance to use it at work yet | count of 15 | count of 10 | Role and employer; whether the chance comes later than day 30 |
| Not ready to use it for real | count of 15 | count of 10 | Exit confidence; whether they practised in front of a mentor |
| Mentor or practice helped | count of 15 | count of 10 | Mentor notes on the same IDs |
Joint display template for the 25 respondents. Write each count against its base; themes can overlap, so columns need not add up.
Guetterman, Fetters and Creswell’s study of joint displays makes the key point: a joint display needs an analytical connection between the columns, not a table of numbers placed beside an unrelated quotation.
How do you read shared themes and contradictions?
In short: A theme in both groups is still evidence, a comment that seems to contradict its score is still data, and a table with no difference is still a result.
If “mentor or practice helped” appears among those who used the skill and those who did not, it may describe a shared experience rather than explain the gap. Learner 0501 said Yes and called the attempt rushed: the score and the words describe different things, and both belong in the record. If no theme differs between the groups, report that, and ask whether the question, the coding or the number of answers could have hidden a difference. Quote supportive and disconfirming examples alike; a vivid quotation illustrates an account, it does not show how common it is.
How do you code without the outcome leading you?
In short: Agree the themes on a varied sample, code the answers without looking at Yes or No, then compare. Check AI-assigned themes against the text.
Reading the No group first primes you to find the explanation you expect. Code blind to the outcome, then bring in the groups. If you use AI, an Intelligence Cell in Sopact Sense reads each open answer as it arrives using a prompt you write, with your theme definitions in it. Inspect the assigned themes beside the original words, especially rare themes and ambiguous answers. If you change the prompt, note the change and the date as a team practice and decide whether earlier answers should be read again under the new wording. For answers in several languages, use the multilingual feedback check before comparing groups.
What can the pattern tell you, and what not?
In short: A reason that appears more often among those who did not use the skill is a lead to test, not a cause you have found.
Everything here is self-reported at one checkpoint, from 25 of 40 completers. The 15 who did not reply may differ from those who did. Role, employer and starting confidence can relate to both the reason and the outcome, and a table does not separate them. A practical step can still be justified without proving the whole pathway. The team writes the finding, the limit and the next step together:
Training example · fictional
“15 of 25 respondents used the skill at work within 30 days; 10 did not, and 15 of 40 completers did not reply. In their own words, those who did not most often described [theme, with count of 10]. This is self-reported and does not establish a cause. We will test a practice session with the next cohort and ask at 30 days whether people practised before using the skill.”
When the next cohort adds a mid-program check, the same linked view carries into pre, mid and post analysis.
How does this work in Sopact Sense?
In short: The score and the open answer arrive on the same ID and wave, so the linked view exists from the first response.
The Intelligence Cell codes each open answer on arrival with your prompt. An Intelligence Row summarizes one person across waves, with scores, their words and mentor notes together. The AI Assistant can then answer “What stopped the completers who did not use the skill?” with every line linked to a record you can open and check. The King Center story describes pre- and post-survey evidence read alongside qualitative feedback across seven programs; the training figures here are fictional.
A 2:34 explainer on why open answers go unread when they are separated from the records they describe, the idea behind the linked view above. Watch on YouTube ↗
ASK ANY TOOL, INCLUDING OURS
On your own data, ask: “For people who answered No at 30 days, show each person’s own reason beside their answer.” A good answer shows one row per person, with the ID, and keeps non-respondents out of the No group. Then ask: “Which reasons appear more often among No than Yes, with counts and bases?” Check that the counts are people, not mentions, and that each theme links to the answers behind it.
Try it on your own data
Open your working evidence plan ↗
- Pick one closed question and the open question that explains it, from the same wave.
- Check both sit on the same ID. Write the linked, unmatched and no-reply counts.
- Write the grouping rule and your themes before reading answers by group.
- Build the joint display with the base for each group on the table.
- Write the finding sentence with its limit and next step, and add the open question to your data dictionary entry as related evidence.
Check your reasoning
For the training team the groups are Yes (15) and No (10), with 15 unknown kept apart. Each theme is counted in people against those bases, so a reason given by some of those who said No reads “n of 10,” never a share of all 40. The finding names the theme and the count, says it is self-reported and not a cause, and points to the practice-session test. If your open question sat in a different tool from the score, your first finding is the unmatched count.
Questions teams ask
Must numbers and comments be on the same row?
They need a reliable link on the same ID and wave. One row per person is the simplest view when one person answers both. When several people describe one person, such as a mentor and the learner, keep separate labelled records on the same ID. For anonymous feedback, connect at the theme or group level instead.
Should outcome groups be defined before coding?
For a planned comparison, yes: write the grouping rule before you read answers by group. Coding without seeing the outcome first helps limit expectation bias. Exploratory comparisons are legitimate too, as long as they are labelled exploratory in the report.
Can comments prove why an outcome changed?
No. They can support or challenge an explanation, and point to what to test next. Check timing, alternative explanations such as role or employer, and other evidence on the same ID. A causal claim needs a design built to support it.
What should I do with unmatched records and non-replies?
Keep them visible. Report the unmatched count and investigate uncertain links rather than forcing a match. Keep people who did not reply as unknown, never as a No, and say how many there are beside every finding.
What if no differences appear between groups?
Report it. Similar experiences, a question that does not ask about the right thing, coarse themes or too few answers can all produce no difference. A sound analysis does not have to find one; it has to show its bases and its limits.