Presenting survey results: from finding to report

How to present survey results: structure of a survey report, methods box, wording a finding, readable tables and charts. With a before-and-after example.

Don't present survey results in the order of the questionnaire. Present them as answers to the questions the decision depends on. A survey report opens with a summary of the three or four most important findings. Each finding states the number, its base (how many responses are behind it, and from whom), a comparison and the uncertainty (how precise the number is). Add a methods box and a section that says what remains open. Tables and charts have to be readable without the surrounding text, and they state the question wording and the number of responses (Jacob et al. 2019, p. 248; Möhring and Schlütz 2019, p. 192).

A finding then looks like this, for example: "71% of customers are satisfied (323 of 455 valid responses). Last year it was 75% (278 of 371). A difference of this size is compatible with the variation between two random samples, so a decline has not been shown." How a sentence like this comes about is the topic of the section How do you word a finding?.

This article answers five questions in turn:

1. Who are you writing for, and how is a survey report structured?
2. Which details about the method does every result need?
3. How do you word a finding, and what does a good table look like?
4. How does a chart show the result without distorting it?
5. How do you say what remains open, and how do you present to different audiences?

How you get to the numbers is described in the article Analyzing online surveys. This article is about the step after that: the analysis is done, and now it has to become a report or a presentation.

In the reports I get to see, a number sometimes appears without its base, a chart exaggerates a small difference, or the most important finding appears only on page 23. This doesn't happen all the time, but it's worth watching out for. The course says why: "Even a methodologically perfect study loses its value if the results are communicated poorly" (Marx 2026, slide 257).

A made-up example runs through the article. In October 2026, a wholesaler of office equipment surveyed its business customers. Of 6,200 active customers, 1,400 were selected at random and invited personally by email, and 463 completed the questionnaire. Management wants to know whether to invest in support for new customers or in complaint handling. I'll use this example to show a sentence, a table and a chart, each first done badly and then done well.

Who are you writing for, and which decision should the report support?

A report is written for the people who make decisions with the results. What they know about methods and what they'll use the report for determines its language, length and structure (Malhotra and Birks 2007, p. 728).

So settle two questions before you write the first sentence:

• Which decision is on the table? Write it down in one sentence. In the example: "We are deciding whether next year's budget goes into support for new customers or into complaint handling."
• Who is reading, and what does that person need? Management needs the answer and how reliable it is, the department concerned needs the details, and an examiner or reviewer needs the complete methods section.

Then you select. From their own practice, Weinreich and von Lindern estimate that about 80% of the analysis work never makes it into the report, because the results play no part in the decision (2008, p. 114). Important individual results go into the appendix in condensed form.

Discuss the most important findings and recommendations with the decision makers before you write the report (Malhotra and Birks 2007, p. 724).

How is a survey report structured?

Malhotra and Birks list eleven elements that most research reports contain, from the letter of transmittal through the executive summary, results and limitations to the appendices. They see the list as a guideline to be adapted to your own project (2007, pp. 725-726; Marx 2026, slide 260). For most surveys in companies, public administration and theses, this can be condensed into three levels.

The summary comes first and is written last. It's often the only part that management reads. It briefly describes the problem, the approach and the procedure, and it gives the most important results, conclusions and recommendations a section of their own (Malhotra and Birks 2007, p. 726; Marx 2026, slide 261). Someone who reads only this page must know the answer to the decision question and how solid it is. What it looks like in the example is shown further down.

The results follow the research questions, not the questionnaire. A questionnaire is ordered for the respondents: easy questions first, demographics at the end. A report is ordered for its readers. Weinreich and von Lindern recommend structuring it by the research questions that were set at the start of the project (2008, p. 115). In the example, the sections aren't called "Question 1 to question 24", but "How satisfied are our customers?", "How satisfied are new customers?" and "How well do we handle complaints?".

The appendix makes the report verifiable. It holds the questionnaire, details on the sample and the fieldwork, and technical details (Malhotra and Birks 2007, pp. 726-728).

Which details about the method does every result need?

A percentage without details about the method can't be put into context. According to Möhring and Schlütz, anyone who publishes a survey should state at least the number of respondents, the sampling method, the response rate (if it can be stated), the survey mode and the survey period, and for representative surveys also the intended population (2019, p. 192). Jacob et al. list almost the same points and add that the label "representative survey" on its own is meaningless (2019, pp. 233 and 247-248).

This gives eight details. The first six appear once in the methods box, the last two with every single result.

How an answer should be read depends on how the question was asked. That's why the wording of the questions with all answer options belongs in the report, and the complete questionnaire in one place (Jacob et al. 2019, pp. 25 and 248).

The base is what's missing most often. In group comparisons, you also need to say how the percentages are calculated: calculate them within each group being compared, as the article on cross-tabulation shows.

The methods box in the example

The box doesn't need more than that. A response rate on its own doesn't tell you whether the people who responded resemble those who were invited. What you can check here is covered in the article on the survey response rate.

How do you word a finding?

A finding is more than a number. For someone to base a decision on it, it needs five building blocks.

1. The number. A proportion, a median or a mean, with no more than one decimal place. Whole numbers are often more appropriate (Möhring and Schlütz 2019, p. 192).
2. The base. How many people responded, and which people? "323 of 455" says more than "71.0%".
3. The comparison. A number on its own is neither high nor low. It only becomes so compared with last year, another group or a target.
4. The meaning. What does it mean for the decision? This is where the number turns into a statement.
5. The uncertainty. How precise is the number, and is the difference larger than you'd expect from a sample anyway?

Base and uncertainty: 71% of whom, and how sure?

"9 out of 10 dentists recommend this toothpaste." Anyone who reads that on a package wants to know whether they can rely on it. Two pieces of information are missing for that.

The first: how many dentists were asked? If it was ten, there are nine people behind the sentence. If just one of them had answered differently, it would be 8 out of 10, so 80% instead of 90%. If it was 1,000, the same 90% has 900 people behind it, and a single answer shifts the result by only 0.1 percentage points. The package carries the same sentence in both cases. But with ten respondents the number hangs on a single answer, and with 1,000 it no longer does.

The second: which dentists are meant? Suppose 100 were asked. 90 didn't want to recommend any particular brand, and of the remaining ten, nine named this toothpaste. Then "9 out of 10" is true, but only for the dentists who recommend a brand at all. Of all 100 dentists asked, it's 9, so 9%.

Your readers are in a similar position with a percentage in your report. When they read "71% are satisfied", they ask the same thing: 71% of how many, and of whom? The answer is the base: the number of responses a percentage refers to, together with the information about who gave them. In the example, that's 455 valid responses from all customers surveyed for overall satisfaction, but only 131 for complaint handling, because that question was put only to customers who had made a complaint. In tables and under charts, the base is usually given briefly as n, for example "n = 455".

The base is also the reason why a number comes with its uncertainty. Not every one of the 6,200 customers was surveyed, only a random sample, and 455 answered this question. If chance had put other customers into the sample, the result could also have come out a little above or a little below 71.0%. As with the dentists: the fewer responses there are behind a number, the more it depends on who happened to be asked. The uncertainty says how much. It gives the range within which the value for the customer group in question is compatible with the data. With 455 responses, it runs from 66.8% to 75.2%, which is 71.0% plus or minus 4.2 percentage points. For the 71 new customers, it runs from 41.9% to 65.1%, which is 53.5% plus or minus 11.6 points. This range is called the 95% confidence interval. How to calculate it is shown in the article P-value and statistical significance, which includes a calculator. The range applies to a random sample like the one in the example and describes only its sampling variation: it doesn't show whether customers who didn't respond would judge differently.

When you compare two numbers, the same question arises for the difference. In the example, 71.0% are satisfied, and last year it was 74.9%. Both numbers come from samples, both vary, so the gap between them varies too. The question is whether 3.9 percentage points is more than this variation alone easily produces. A test answers that, and its result is the p-value. Here, p = 0.21: if nothing had changed among the customers, a gap at least this large would still turn up in about 21 out of 100 such comparisons. That happens often, so a decline has not been shown. Only when the p-value is below the usual threshold of 0.05, that is, fewer than 5 out of 100, do we call a difference statistically significant (P-value and statistical significance). Make clear which differences are significant and which are not (Weinreich and von Lindern 2008, p. 109).

Two things follow for your report. With every percentage, state the base: how many responses, and from whom. And with every number a decision rests on, state the uncertainty: for a single number the range, for a comparison whether the difference is significant.

In short: the base says how many responses are behind a number and who gave them. The smaller the base, the less precise the number usually is. The uncertainty says how imprecise: as a range around the number and, in a comparison, as a statement of whether the difference is larger than the variation between samples would lead you to expect.

Before and after: a sentence

In the example, 323 of 455 customers rated their overall satisfaction 4 or 5, which is 71.0%. Last year it was 278 of 371, so 74.9%, measured in a separate random sample with the same question.

Before:

Customer satisfaction has fallen markedly compared with last year (70.99% instead of 74.9%).

The sentence gives a number and a comparison. What's missing is the base, what "satisfied" means, and the uncertainty. The word "markedly" is a judgment that the data don't support. And two decimal places suggest a precision that 455 responses can't deliver.

After:

71.0% of customers are satisfied or very satisfied with us (323 of 455 valid responses, ratings of 4 or 5 on a five-point scale). Last year, the figure was 74.9% (278 of 371). The decline of 3.9 percentage points is not statistically significant (p = 0.21): a difference of this size is compatible with the sampling variation between two random samples. The 95% confidence interval for the change ranges from a decline of 10.0 percentage points to an increase of 2.1 percentage points. A decline has therefore not been shown. We will reassess this question in the next survey.

This finding is longer, but every detail has a job. "Not significant" doesn't mean "no difference", and it doesn't mean "chance" either. It means that the survey neither demonstrates a change nor rules one out. The interval shows what remains compatible with the data. More on this in the section Comparing two groups or two years.

For the summary, the finding can be shortened without becoming wrong: "71% of customers are satisfied (n = 455). A change from last year has not been shown."

What does a good table look like?

Not every table a program puts out is suitable for a report (Jacob et al. 2019, p. 271). Five rules follow from Malhotra and Birks (2007, pp. 730-731; Marx 2026, slide 264), Jacob et al. (2019, pp. 248-249) and Möhring and Schlütz (2019, p. 192):

• Number and title. The title says briefly what the table shows. You refer to the table in the text by its number.
• Question wording. Below the title or in the footnote.
• Base in a row of its own. The number of responses per column, plus a note on whether column, row or total percentages are shown.
• Ordered by what matters. Time series by time, categories without a natural order by size. The points of a scale stay in their order.
• Not overloaded, readable on its own. Percentages with no more than one decimal place, whole numbers for small groups. For comparisons, the statistical measures go in the footnote.

Before and after: a table

Before:

What goes wrong here: the title and the question are missing. What does "satisfaction" mean: a mean, the top scale point or the top two? The base is missing, even though there are only 71 responses behind the new customers. There, a single person shifts the proportion by 1.4 percentage points, so the second decimal place is meaningless. And the last column belongs to a different question with a different base: only the 131 customers who had made a complaint in the past twelve months were asked about complaint handling.

After:

Table 1: Overall satisfaction by customer group

Question: "Overall, how satisfied are you with us as a supplier?" Scale from 1 (very dissatisfied) to 5 (very satisfied).

Column percentages, rounded. 8 of the 463 respondents did not answer the question. Difference in the proportion of satisfied customers between existing and new customers: 20.7 percentage points (from the unrounded proportions 74.2% and 53.5%), 95% confidence interval 8.3 to 33.1 points, p < 0.001 (test for two proportions).

The table now answers one question: are new customers less satisfied? Yes. The difference is significant, but because the group of new customers is small, the estimate of its size is imprecise: the interval runs from a little over 8 to a little over 33 percentage points. Complaint handling gets a table of its own with its own base: 55 of 131 customers who made a complaint, so 42.0%, are satisfied.

How does a chart show the result without distorting it?

The choice of display and the scaling can change the message of a chart considerably (Weinreich and von Lindern 2008, p. 111). The form follows the content: bars and columns for comparisons, pies for shares of a whole, lines for trends (Möhring and Schlütz 2019, p. 192; Malhotra and Birks 2007, pp. 731-732). For the six most common cases in a survey, this means:

A survey report rarely needs more than that. Other chart types will be covered in a separate article.

Five typical mistakes

1. The truncated axis. If the axis of a column or bar chart doesn't start at zero, small differences grow visually. Weinreich and von Lindern show this with a column chart: the more of the scale is cut off, the more differences are overestimated, and some programs cut it off on their own (2008, pp. 113-114). In the example: if the axis starts at 68%, the column for 2025 with 74.9% is about 2.3 times as tall as the one for 2026 with 71.0%. On an axis starting at zero, it's 1.06 times as tall.

2. Percentages without a base. "42.0% are satisfied with complaint handling" sounds like all customers. What's meant is 55 of the 131 customers who had made a complaint. Relative to all 463 respondents it would be 11.9%, and that number would be just as misleading, because the other 332 weren't asked at all. That's why the question and the number of responses belong under every chart (Möhring and Schlütz 2019, p. 192).

3. Pie and 3D in the wrong place. A pie chart shows simple shares of a whole. It's no good for trends or for relationships between several variables (Malhotra and Birks 2007, p. 731). For a multiple-choice question it's wrong: in the example, 73.0% use the online shop, 44.7% email, 34.8% the phone and 14.3% the field sales team (proportion of the 463 respondents, multiple answers possible). The total is about 167%, and a pie would silently turn that into 100%. On 3D effects, the sources differ in how strict they are: Malhotra and Birks urge great caution, because the segments appear distorted (2007, pp. 731-732), Möhring and Schlütz advise using them sparingly (2019, p. 192), and Linke recommends charts without 3D (2018, p. 104).

4. Too many categories. There's no fixed upper limit. As a rule of thumb, Malhotra and Birks give a maximum of seven segments (2007, p. 731; Marx 2026, slide 266), and in the article on analysis I advise five to six. The reason is the same: with every additional segment the chart gets harder to read, until it no longer shows the data but hides them. If you have considerably more categories, the pie chart is probably not the right form. A bar chart shows the same shares more legibly. Stacked and grouped bars also work only with a few values (Malhotra and Birks 2007, p. 732). Combine categories with few responses.

5. The wrong order. Data are ordered by what matters most about them (Malhotra and Birks 2007, p. 730). For categories without a natural order, such as ordering channels, that's size. The points of a rating scale keep their order.

Before and after: a chart

Before, in words: a column chart titled "Customer satisfaction", two columns in 3D, the axis from 68% to 76%, no values, no number of responses. The impression: satisfaction has fallen to less than half.

After, in words: the title is the finding. For each group there's a dot on an axis from 0 to 100%, a line shows the 95% confidence interval, the number of responses is on the right, and the question wording is below. The chart shows only this one question.

The intervals for 2025 and 2026 overlap over more than half of their width. The result for the new customers is far less precise than the one for the existing customers, because there are only 71 responses behind it. But whether a difference is significant isn't decided by how overlapping intervals look. It's decided by the test of the difference. The intervals also describe only the sampling variation. They don't say whether the 463 people who responded differ from the other 937 who were invited.

Complaint handling is missing from this chart on purpose: it's a different question with its own base (55 of 131, 95% confidence interval 33.5% to 50.4%).

Small groups and anonymity in the report

The more finely you break results down, the smaller the groups get. That creates two problems.

The first is how much the numbers can tell you. As a rule of thumb, subgroups should contain at least 20 to 25 cases (Jacob et al. 2019, p. 72). Weinreich and von Lindern show a chart in which one age group looks considerably less satisfied than the next, except that there are 23 respondents behind one and 3 behind the other (2008, pp. 109-110). Don't report such groups separately, or combine them.

The second is protecting the respondents: in small groups, answers can be traced back to individual people. The article Employee survey explains which minimum group sizes are common and why no threshold protects against individual answers being worked out by subtraction. The legal framework is described in Data protection in online surveys.

Set the minimum size before you analyze, and write it into the methods box. In the example it's 30 responses.

What remains open? Naming limitations honestly

Every survey has limitations. Malhotra and Birks call for a section of its own for them, and it has two jobs: decision makers shouldn't rely too heavily on the results or apply them to groups they don't hold for, and confidence in the survey shouldn't suffer without reason (2007, p. 728).

The opposite is tempting: shaping ambiguous results into a neat story instead of admitting that the data give no clear answer (Day 1974, cited in Malhotra and Birks 2007, p. 737; Marx 2026, slide 270).

Something can remain open in three ways:

The third point includes the type of selection. The tests and confidence intervals used here require random selection (Jacob et al. 2019, p. 249). They don't capture bias from nonresponse. If your responses come from an open link, report tendencies and describe how the participants were recruited (Wording for your methods section). Label comparisons that weren't planned as exploratory: if you test many groups, you'll find a few significant differences for that reason alone.

In the example, this makes the recommendation more cautious and more useful at the same time. For the new customers, the gap is established. For complaints, only 55 of 131 customers are satisfied with how they were handled, and no comparison figure is available yet. Which investment pays off more is something the survey doesn't answer. The report says so and proposes the next step.

The summary in the example

Now, at the end, the first page can be written:

How do you present the same results to different audiences?

If a report has several groups of readers, separate sections for each group or separate versions help (Malhotra and Birks 2007, p. 729). The results stay the same. What changes is the selection and the depth.

Many managers form their impression of a project during the presentation. Tailor it to the audience and rehearse it (Malhotra and Birks 2007, p. 733; Marx 2026, slide 267). Weinreich and von Lindern give practical advice on this (2008, pp. 116-117):

• Show a few solid results and keep slides with details ready for follow-up questions.
• Use charts and text large enough to read, as a rule with a font size of at least 14 points.
• Plan no more than half of the time for the talk. The rest belongs to the discussion.
• Talk to the client before the meeting, because priorities may have changed.

Every slide has a finding as its title, and below it a chart or a table with the base. If a question comes up that the survey doesn't answer, say so and suggest how it could be clarified (Weinreich and von Lindern 2008, p. 117).

Reports and charts with QUESTIONSTAR

In QUESTIONSTAR, the first analysis report is generated automatically. It contains descriptive statistics for all questions and updates itself when you edit the questionnaire (update note). By default, only completed questionnaires go into the reports. You create additional reports for focused views (Features; video tutorial on reports).

What you can use for the steps in this article:

• Tables and charts: For every question you see the count, percentage and cumulative percentage for each answer. Descriptive statistics can be switched on in practically every report, including for a pie chart and for single-choice as well as multiple-choice questions. They show the mean, standard deviation, variance, minimum, maximum and quartiles. The available chart types are pie, bar, column, line, histogram, box, scatter, radar and treemap, with stacking by values or by shares. You change the display type per question.
• Base for multiple-choice questions: What the percentages refer to is a setting: the respondents who answered the question, or all answers given. For the report you usually need the first option, the proportion of people.
• Grouping and hiding answers: This is how you combine categories with few responses before you pass a chart on (video tutorial, from minute 1:57).
• Filters and crosstabs: With filters you analyze one group specifically. The "Crosstab" report compares groups with column, row and total percentages and calculates the chi-square test.
• Export: Individual charts and tables can be exported on their own, and the whole report to Word, Excel, PowerPoint and PDF, according to the feature overview "with real, editable Office charts, not images".
• Sharing a report: Reports can be published online and shared via a link, without the recipients having to log in.
• Questionnaire for the appendix: You export the questionnaire as a Word or PDF document, with question numbers if you like (Help: Export questionnaire).
• Rehearsing before fieldwork: With generated test data you build the report before the first real response comes in, and you notice if a question is missing.

Three tasks remain your responsibility. The reports don't show confidence intervals: you calculate the margin of error of a proportion with the confidence interval calculator. You code open-ended responses outside the tool. And QUESTIONSTAR doesn't hide small groups on its own: if "Anonymize IDs" is active in the privacy settings, reports appear only once ten questionnaires have been fully completed (Help: Privacy), and you have to enforce your minimum size per group yourself, using filters and access rights. Before sharing and after every export, check the number of responses, filters, axes and labels.

Checklist: reporting survey results

• The decision the report is meant to support is written down in one sentence
• Groups of readers named, version per group decided
• One-page summary: findings with number and base, recommendation, most important limitation
• Methods box: population, selection, period and mode, number of respondents, response rate with denominator, weighting, minimum size for groups
• Results ordered by research questions, not by the questionnaire
• Every finding with number, base, comparison, meaning and uncertainty, "not significant" not worded as "no difference"
• Every table with number, title, question wording, base row and direction of the percentages, percentages with no more than one decimal place
• Charts: bars and columns for proportions with an axis starting at zero, no 3D, no pie for multiple-choice questions, a pie only with few categories, question and n below
• Section "Limitations and open questions" written: not asked, not shown, not transferable
• Questionnaire in its exact wording in the appendix, report proofread by two or three people who don't know the project

Frequently asked questions about presenting survey results

What is the best way to present survey results?

Present survey results ordered by the questions the decision depends on, not by the questionnaire. Each finding states the number, the base, a comparison and the uncertainty. Tables and charts state the question and the number of responses (Jacob et al. 2019, p. 248; Möhring and Schlütz 2019, p. 192).

How is a survey report structured?

A survey report opens with a summary of the most important findings and the recommendation. Then come the details about the method, the results for each research question, the conclusions, a section on limitations and open questions, and the appendix with the questionnaire and tables (based on Malhotra and Birks 2007, pp. 725-726).

Which details about the method belong in every survey report?

Every survey report should state the population, the sampling method, the period, the survey mode, the number of respondents, the response rate with its denominator and, where applicable, the weighting. In addition, each result needs the question wording and the base of the percentage (Möhring and Schlütz 2019, p. 192; Jacob et al. 2019, pp. 233 and 248).

Which mistakes make charts in survey reports lose credibility?

Five above all: truncated axes in bar and column charts, percentages without a base, pie charts for multiple-choice questions and 3D effects, too many categories, and scale points in the wrong order. For pie charts, five to seven segments is a rule of thumb, not a fixed limit (Malhotra and Birks 2007, p. 731; Weinreich and von Lindern 2008, pp. 113-114; Möhring and Schlütz 2019, p. 192).

Do I have to report results that are not significant?

Yes, if they belong to the decision question. Report the difference with its size and the confidence interval, and state that a real difference has not been shown. Leaving out results because they don't fit the picture violates the objectivity a report needs (Malhotra and Birks 2007, pp. 729 and 736-737).

How do you present survey results to management?

Start with the answer to the decision question, then three or four findings, one per slide with a chart and the base, then the recommendation and the limitations. Plan no more than half of the time for the talk (Weinreich and von Lindern 2008, pp. 116-117).

Sources

Jacob, R., Heinz, A. and Décieux, J. P. (2019). Umfrage. Einführung in die Methoden der Umfrageforschung [Surveys: an introduction to the methods of survey research] (4th ed.). Berlin/Boston: De Gruyter Oldenbourg. In German. p. 25 (documenting the question wording), p. 72 (number of cases per subgroup), pp. 232-233 (methods report, minimum details when publishing), pp. 247-249 (checklist for the research report, tables, significance only with random selection), p. 271 (tables).

Linke, R. (2018). Mitarbeiterbefragungen optimieren. Von der Befragung zum wirksamen Management-Instrument [Optimizing employee surveys: from the survey to an effective management tool]. Wiesbaden: Springer Gabler. In German. DOI 10.1007/978-3-658-17722-5. pp. 104-105 (clear presentation of results, charts without 3D, showing the distribution).

Malhotra, N. K. and Birks, D. F. (2007). Marketing Research. An Applied Approach (3rd European edition). Harlow: Pearson Education. Chapter 25, pp. 722-740: p. 724 (discussing findings in advance), pp. 725-728 (report format, executive summary, results, limitations, appendix), pp. 728-729 (readers, objectivity), pp. 730-732 (tables and graphs), p. 733 (oral presentation), pp. 736-737 (integrity, cited there: Day, R. L. (1974). A comment on "Ethics in marketing research". Journal of Marketing Research, 11, 232-233).

Marx, P. (2026). Principles of Survey Research. Online course, chapter 7 "Reporting Results", slides 257, 260, 261, 264, 266, 267 and 270. QUESTIONSTAR. Slide 257, slide 260, slide 261, slide 264, slide 266, slide 267, slide 270.

Möhring, W. and Schlütz, D. (2019). Die Befragung in der Medien- und Kommunikationswissenschaft. Eine praxisorientierte Einführung [Surveys in media and communication studies: a practical introduction] (3rd ed.). Wiesbaden: Springer VS. In German. DOI 10.1007/978-3-658-25865-8. p. 192 (presentation of the data, tables, base, chart form, details about the method).

Schumann, S. (2019). Repräsentative Umfrage. Praxisorientierte Einführung in empirische Methoden und statistische Analyseverfahren [Representative surveys: a practical introduction to empirical methods and statistical analysis] (7th ed.). Berlin/Boston: De Gruyter Oldenbourg. In German. p. 164 (bar chart and pie chart).

Weinreich, U. and von Lindern, E. (2008). Praxisbuch Kundenbefragungen [Customer surveys: a practical handbook]. München: mi-Fachverlag, FinanzBuch Verlag. In German. pp. 109-110 (assessing results, small groups), pp. 111-114 (graphical presentation, scaling), pp. 114-116 (results report, checklist, sample outline), pp. 116-117 (presenting results).

I wish you reports that get read, and readers who finish them knowing what to do next and which questions remain open.

Related articles:

• Analyzing online surveys: it starts before you collect data
• P-value and statistical significance, explained simply
• Cross-tabulation: how to read and test a contingency table

Dr. Paul Marx

Author: Dr. Paul Marx
Dr. Paul Marx is the founder and Managing Director of QUESTIONSTAR and was a professor of marketing at the University of Siegen.