"Xiao Zhang, how did you analyze the multiple-choice data in this survey?" "I... I just counted how many times each option was selected." "Then how do you tell apart the people who selected both A and B from those who selected only A?" Xiao Zhang went silent.
This is a real dilemma for many people: when designing a survey, multiple-choice questions feel information-rich and broad in coverage, but at the analysis stage you realize — single-choice data can be converted to percentages, while multiple-choice data has a completely different structure that traditional analysis methods simply don't apply to. Today, let's systematically break down the principles, operations, and interpretation of multiple-response analysis.
First, Understand: Why Is the Data Structure of Multiple-Choice Questions "Special"?
1.1 Single-Choice vs Multiple-Choice: How Data Storage Differs
A single-choice question (e.g., "Your gender?"): one variable, one value.
| Respondent | Gender |
|---|---|
| 1 | Male |
| 2 | Female |
| 3 | Male |
A multiple-choice question (e.g., "Through which channels did you learn about us?"): one question, N options, and each option is an independent variable. In a multiple-choice question, each respondent can "check" multiple options, so the data is binary (selected = 1, not selected = 0):
| Respondent | Douyin | Xiaohongshu | ||
|---|---|---|---|---|
| 1 | 1 | 1 | 0 | 0 |
| 2 | 1 | 0 | 1 | 0 |
| 3 | 0 | 1 | 1 | 1 |
1.2 Why Can't You Just Analyze by Counting?
A common mistake is to simply add up the selection counts of each option and rank them by frequency. The problems with doing this:
- It ignores the relationships between options (are people who chose A also more likely to choose B?);
- It cannot perform cross-tabulation (how do the choice patterns of men and women differ?);
- It cannot perform statistical testing (is the difference real, or just random fluctuation?).
The professional method: Multiple Response Analysis.
The Two Core Methods of Multiple-Response Analysis
2.1 Method 1: Frequency Analysis
Principle: Count the number and proportion of times each option is selected.
Response percentage = number of times an option was selected ÷ total number of selections across all options (the option's share of "all selections");
Case percentage = number of times an option was selected ÷ total number of respondents (the proportion of respondents who selected that option).
Working example: Survey 200 people, asking "Which social media platforms do you use?" (multiple choice):
| Option | Times Selected | Response Percentage | Case Percentage |
|---|---|---|---|
| 180 | 32.7% | 90.0% | |
| Douyin | 150 | 27.3% | 75.0% |
| Xiaohongshu | 120 | 21.8% | 60.0% |
| 100 | 18.2% | 50.0% | |
| Total | 550 | 100% | — |
Interpretation: Response percentage — among all 550 selections, WeChat accounted for 32.7%; case percentage — 90% of respondents use WeChat. Pro tip: response percentages sum to 100%, while case percentages usually exceed 100% (because each person selected multiple options).
2.2 Method 2: Cross-Tabulation (Group Comparison)
Principle: Compare the differences in option selection across different groups and run a chi-square test. Working example: Compare the social media usage preferences of "male" and "female" respondents:
| Option | Male (n=100) | Female (n=100) | Chi-Square Test |
|---|---|---|---|
| 88% | 92% | p>0.05, no significant difference | |
| Douyin | 70% | 80% | p<0.05, significantly higher among females |
| Xiaohongshu | 30% | 90% | p<0.001, significantly higher among females |
| 55% | 45% | p>0.05, no significant difference |
Interpretation: Douyin and Xiaohongshu show significant gender differences, with Xiaohongshu showing the largest gap (30% vs 90%). Advanced analysis: You can also analyze "option combinations" — for example, what proportion of people "use both WeChat and Douyin"? What characteristics do these people share?
SPSS Step-by-Step Guide
If you're analyzing with SPSS, here is the standard workflow:
- Step 1: Define the multiple-response set. Path: Analyze → Multiple Response → Define Variable Sets. Move all option variables of the multiple-choice question into "Variables in Set", set "Variables Are Coded As" to "Dichotomies" with a counted value of "1", and give the set a name (e.g., "Social Media").
- Step 2: Frequency analysis. Path: Analyze → Multiple Response → Frequencies. Select the multiple-response set you defined to output response percentages, case percentages, and the valid sample size.
- Step 3: Cross-tabulation. Path: Analyze → Multiple Response → Crosstabs. Put the multiple-response set in rows and a grouping variable (e.g., gender) in columns, and optionally request row/column percentages and the chi-square test.
Advanced Analysis: From Description to Insight
- 4.1 Association rule analysis (market basket analysis): Mine which options tend to "appear together". Example: A survey asking "Which factors matter to you when buying a phone?" found that "battery life" and "fast charging" are strongly associated — of the people who chose battery, 80% also chose fast charging. Marketing implication: these two selling points should be presented together in product promotion.
- 4.2 Correspondence analysis: Map "respondent groups" and "options" onto a two-dimensional space simultaneously to visually show "which type of person tends to choose which option". Example: "College students" are positioned near "low price" and "stylish appearance", while "working professionals" are positioned near "brand recognition" and "after-sales service".
- 4.3 Latent class analysis (LCA): Based on multiple-choice response patterns, automatically divide respondents into different "types". Example: In a survey on "your leisure activities" (watching movies, sports, gaming, travel, reading...), LCA identified three groups — homebodies (gaming + movies), socializers (travel + gatherings + sports), and the literary type (reading + movies + exhibitions). The marketing strategies for these different types are completely different.
Real-World Case: How an Education Institution Used Multiple-Choice Analysis to Optimize Course Design
Background: An online education institution wanted to understand "which factors learners care about when choosing a course", so it designed a multiple-choice question (limited to 3 selections) and collected 500 valid responses.
Step 1: Frequency analysis
| Option | Case Percentage |
|---|---|
| Faculty quality | 78% |
| Learner reviews | 65% |
| Trial lesson experience | 58% |
| Certificate value | 45% |
| Course price | 42% |
| Brand recognition | 30% |
| Course duration | 22% |
Initial insight: faculty quality, learner reviews, and trial lesson experience are the top three.
Step 2: Cross-tabulation (by age group)
| Option | 18–25 | 26–35 | 36+ |
|---|---|---|---|
| Faculty quality | 65% | 82% | 88% |
| Course price | 58% | 38% | 25% |
| Certificate value | 32% | 52% | 48% |
Deeper insight: younger groups are more price-sensitive, while older groups place more weight on faculty quality and certificate value.
Step 3: Association rule analysis. It found that "learner reviews" and "trial lesson experience" are strongly associated (lift = 2.8). Marketing implication: promote free trial lessons together with positive learner reviews, and adjust the promotional focus for different age groups.
Business outcome: The course detail page was redesigned to highlight faculty + trial lessons + learner reviews, and a "first-order discount" label was added for younger groups, boosting the conversion rate by 22%.
Final Thoughts
Multiple-choice questions are the "information goldmine" of survey research — they reflect the complexity of the real world better than single-choice questions. But a goldmine needs the right mining methods; otherwise it's just ore piled up in a warehouse. Multiple-response analysis is the tool for mining this goldmine. From basic frequency analysis to cross-group comparison, and further to association rule mining and latent class analysis, every level yields different insights. Want to complete all of these analyses in one click? The TLSUN survey system can directly generate cross-tabulation reports and export them in SPSS format — welcome to try it.