Cross tabulation analysis of survey results: examples and best practices

Cross tabulation, often shortened to cross-tab, is a method of organizing survey data into a table that shows how two or more variables relate to each other. Instead of reporting each question's results in isolation, a cross tabulation table places one variable along the rows and another along the columns, so the overlap between them is readable at a glance.

A lot of teams stop after looking at the overall satisfaction score. If it's reasonably high, the assumption is that everything is on track. That isn't always the case. We've seen situations where the average looked perfectly healthy until the results were broken down by customer type. Existing customers were happy, newer ones clearly weren't. The overall score never showed that because both groups had been averaged together.

It's the same underlying logic used in epidemiology to compare treatment against outcome, in political polling to compare demographic groups against candidate preference, and in market research to compare purchase behavior against income bracket.

Here is a detailed breakdown of the different steps involved in cross tabulation analysis.

• Start with the question you're trying to answer. Before pairing survey questions together, think about what you're actually trying to learn. Not every combination will tell you something useful. If customer churn is increasing, for example, it makes more sense to compare likelihood to renew with support ticket volume or whether customers completed onboarding than with something unrelated like their preferred contact method. The business question should always guide the comparison.
• Keep the layout consistent. Most analysts place the variable they think is influencing the outcome in the rows and the outcome itself in the columns. For example, customer tenure might appear in the rows while satisfaction scores sit across the top. There's no strict rule that says you have to build the table this way, but using the same layout throughout a report makes the results much easier for other people to follow.
• Build the table and convert counts to percentages. Each cell shows the count of respondents in that combination of row and column categories. A basic 2x2 table might show gender against product preference, while a more detailed one shows five age brackets against five satisfaction levels (a 5x5 grid). Raw counts only work if subgroups are the same size. 40 out of 50 satisfied enterprise customers is a higher raw count than 120 out of 200 small-business customers, but a lower percentage (80% versus 60%). Row or column percentages fix this and are usually what should be presented.
• Make sure the difference actually means something. It's easy to compare two groups and assume you've found the answer. Maybe enterprise customers look happier than small businesses, or newer users seem less satisfied than long-time customers. Sometimes those differences are real. Sometimes they're just the result of the particular group of people who happened to answer your survey. Before sharing the results, it's worth checking whether the pattern is likely to hold up with a larger audience. A chi-square test is commonly used for that. If some categories contain only a small number of responses, the results become less dependable. In those cases, analysts often combine categories or use Fisher's exact test instead.

Consider this business case.

One SaaS company reviewed its latest NPS survey and saw almost no movement in the overall score. At first, it looked like nothing had really changed. Looking a little closer told a different story. Customers on the highest-priced plan were overwhelmingly promoters, while those on the entry-level plan were much less satisfied. The average score had blended those groups together, making the differences easy to miss. Breaking the results down by subscription plan gave the team a much clearer idea of where to focus next, whether that meant improving onboarding, reviewing feature access, or looking more closely at the customer experience.

Building these tables by hand in a spreadsheet every reporting cycle gets old fast, and it's easy to mislabel a row or misplace a percentage. Zoho Survey's cross-tab reporting tool builds this into the reporting workspace. So, the table updates as new responses come in.

Building a cross-tab report in Zoho Survey starts with the questions you've already asked. You can compare multiple-choice, dropdown, rating scale, NPS, ranking, or matrix questions without creating new fields or reorganizing your survey. If you want to look at more than one comparison at a time, you can also place several variables side by side. For example, you could see customer satisfaction broken down by both region and subscription plan in the same report instead of generating separate tables.

Incomplete responses can sometimes throw off the percentages, especially if a noticeable number of respondents skipped a question. The report lets you decide how those skipped answers should be handled. It can also be shared on a schedule, which is useful when the same analysis is reviewed every month.

None of this replaces the thinking part. You still decide which variables matter. It just removes the manual table-building that used to eat into interpretation time.

You don't need a cross-tab for every survey question. In most cases, a few well-chosen comparisons will tell you far more than dozens of tables that never get used.

Watch your subgroup sizes before trusting a percentage. Pew Research Center's own methodology notes give a useful illustration: in a simple random sample of about 1,067 people, a subgroup such as Hispanic adults, who make up roughly 15% of the U.S. adult population, would be represented by only around 160 cases if sampled proportionately, pushing that subgroup's margin of error to roughly plus or minus 8 percentage points on its own, and to roughly plus or minus 16 points when comparing two such subgroups against each other. A cell with a dozen responses can look dramatic and mean almost nothing statistically.

It's easy to build a report that's far more detailed than anyone actually needs. In most cases, a few well-chosen comparisons will answer the question much better than pages of tables. Percentages also tend to be easier to compare than raw numbers, especially when one customer group is much larger than another. And if a result looks surprising, it's worth checking that it's based on enough responses before treating it as a real pattern.

Think about the audience as well. Someone working with the data every day may want to see every category. But an executive presentation usually works better when similar groups have been combined into something easier to scan. Simple labels also go a long way. A report should make sense even if the person reading it wasn't involved in creating it.

Don't think of cross-tabulation as something you do once and forget about. The real value often comes from repeating the same analysis over time. That's when you start to see whether the differences between customer groups are changing or staying the same, which is often the information teams end up acting on.

It's easy to look at a cross-tab and assume you've found the reason something is happening. In reality, all you've found is that two things appear to be related. For example, customers on annual plans may report higher satisfaction than those on monthly plans. That doesn't necessarily mean annual billing makes people happier. It could simply mean that satisfied customers are more willing to commit to a longer subscription. The table points you toward a question, but it doesn't answer it.

Another issue comes from breaking the data into too many small groups. It can be useful to compare results by region or customer segment, but adding more and more categories quickly reduces the number of responses in each group. Before long, you're trying to draw conclusions from only a handful of people. At that point, the findings become much less reliable. In most cases, a few well-chosen comparisons will tell you far more than splitting the data into every combination possible.