Employee survey: process, questions and analysis

How to plan, run and analyze an anonymous employee survey: the process in eight steps, sample questions, pulse surveys and actions. With templates.

An employee survey is a systematic, usually voluntary and anonymous survey of a company's employees with a standardized questionnaire. It shows how the workforce experiences work, leadership and collaboration, and where things get stuck. It becomes useful under two conditions. Employees must be able to answer without fearing negative consequences. And afterward they must see that something happens with the results.

The example is a German company, so works councils and German law appear throughout. Sabine Albers heads HR at a mid-sized machine manufacturer with 240 employees. 150 work in production, many of them in shifts and without a computer of their own, and 90 in administration. Management wants to know "how satisfied people are". Ms. Albers suspects that this question alone won't get her very far. The company, people and numbers in this example are made up.

What is an employee survey?

Employee surveys can be designed in different ways (Wiertalla 2013, p. 96). Typical features are: everyone or a sample is surveyed, participation is voluntary, anonymity is guaranteed and the questionnaire is standardized. Added to that are prompt feedback and repetition at regular intervals (Borg 2003 and Bungard et al. 2007, cited in Werther 2015, pp. 27-28).

Company management wants to use it, for example, to identify strengths and weaknesses or to check whether actions have worked (Nürnberg 2017, p. 20). Employees and the works council, the body that represents employees in a German company, want to help shape working conditions, for example (Nürnberg 2017, p. 21). By the way, 360-degree feedback is something else. It's about one individual person.

Survey everyone or just a sample?

In practice, a sample is usually not enough for results by department, because it hardly covers small units (Nürnberg 2017, p. 50; Linke 2018, p. 34). If Ms. Albers randomly draws 60 of 240 employees, then on average only 60 × 24 / 240 = 6 people from a department of 24 end up in the sample. That's very few for reliable group comparisons. Whether a group reaches the agreed threshold for its own report depends on how many people actually respond. So in a company of this size, I advise inviting everyone. If you do want a sample, the sample size calculator helps.

Which topics belong in it?

Factors that influence job satisfaction include work equipment and working conditions, development, pay, team climate, direct leadership, information and commitment (Borg 2003, cited in Linke 2018, pp. 75-76). A practitioner's book recommends no more than ten topic areas (Nürnberg 2017, p. 26). Only ask about things the company can and wants to change, otherwise you create dissatisfaction (Linke 2018, p. 80). But if you leave out an important topic like pay entirely, you get a sugarcoated picture.

Satisfaction or engagement?

Job satisfaction is an attitude toward one's own work. A classic definition describes it as a pleasant, positive emotional state that results from appraising one's own work (Locke 1976, p. 1300, cited in Spichalsky 2016, p. 38). Work engagement is a positive, fulfilling state characterized by vigor, dedication and absorption, that is, being completely immersed in one's work (Schaufeli et al. 2002, p. 74). Put simply, satisfaction describes how someone feels about their work, and engagement how much energy they bring to it.

Every instrument defines these terms a little differently. With its Q12, Gallup measures emotional commitment to the employer (Linke 2018, p. 76). According to Gallup's German index, the Engagement Index Deutschland 2025, 10% of employees in Germany had high emotional commitment to their employer (Gallup 2026). However, external standard models assume that your employees have the same needs as the people on whom the model was tested (Linke 2018, p. 7).

Measuring employee satisfaction

You measure employee satisfaction with an anonymous, standardized survey: one overall question on satisfaction plus statements on influencing factors such as leadership, working conditions and information. Key figures show you where to look more closely, but not why.

Staff turnover and sick leave are suitable as comparison figures, because they're usually already available by department (Linke 2018, p. 131). Research has examined whether lower job satisfaction goes along with more absenteeism and stronger intentions to quit (Spichalsky 2016, pp. 36-37). But these figures have many causes (Linke 2018, p. 132).

At Ms. Albers' company, sick leave in assembly has been rising for a year. Is it the workload, the shift schedules or a flu wave? That's why combining the survey with key figures is recommended (Levenson 2014, cited in Nürnberg 2017, p. 26).

Why not just ask in a conversation?

For honest criticism, I think a face-to-face conversation is the wrong route. Who likes to openly criticize their own boss? If you simply ask directly, you often measure the Monday-morning mood or what the person wants their boss to believe (Linke 2018, p. 8). In a written anonymous survey, on the other hand, interviewer effects disappear (Jacob et al. 2019, p. 118). I don't know of any study that compares the two directly within a company. But a lot speaks in favor of the anonymous survey.

Is one question about satisfaction enough?

"How satisfied are you overall?" can be more useful than you might expect. A single question about job satisfaction gives results close to those of a more detailed scale (Werther 2015, p. 42). Put it at the beginning, before individual questions influence the overall judgment. A methods chapter gives the reasoning for customer surveys (Thielsch et al. 2012, p. 236). It just doesn't tell you anything about the why (Linke 2018, pp. 73-74).

Even a good question misleads if you only look at the mean. Two teams of 20 people each answer "Problems are raised and resolved promptly here" on a scale from 1 (strongly disagree) to 5 (strongly agree):

Both are at 3.0. Team A is largely of one mind, Team B is split into two camps. That's why the distribution belongs next to the mean (Linke 2018, p. 105).

The employee survey process in eight steps

A textbook divides a survey into the phases of decision, planning, communication, survey, analysis, feedback and evaluation (Werther 2015, p. 40). I turn them into eight practical steps.

1. Clarify the goal, win over management. Write down the goals in a few sentences and have management approve them (Nürnberg 2017, p. 27). Without management's visible commitment, a survey brings little benefit (Linke 2018, p. 68).
2. Involve employee representatives and data protection. Where applicable, bring in the works council or staff council and the data protection officer from the start (Werther 2015, p. 40). Decide in advance which analyses are permitted and who sees which results (Werther 2015, p. 31). More on this
3. Create and test the questionnaire. Test it with the most important groups, such as production and administration (Linke 2018, p. 35), and measure how long it takes (Nürnberg 2017, p. 41). A written survey shouldn't take longer than 20 minutes (Jacob et al. 2019, p. 118). Sample questions
4. Inform in advance. Say why you're asking, whom you're asking and what you're asking, and which follow-up actions are planned (Nürnberg 2017, p. 55). If the works council recommends taking part, that has more effect in some groups than an appeal from management (Linke 2018, p. 62).
5. Run the survey. Plan around vacations, school holidays, public holidays and peak periods (Linke 2018, pp. 30-31). An online survey with one or two reminders can be completed in two to three weeks (Jacob et al. 2019, p. 121). That's a general guideline.
6. Analyze. You can often work with the results two to three weeks after the survey (Linke 2018, p. 31). Analysis
7. Report back. Thank employees as soon as the survey closes so they are not left without an update (Linke 2018, pp. 58-59). The key results should be available within the first two months (Nürnberg 2017, p. 71). Agree with the works council when and how it receives them (Linke 2018, p. 71).
8. Implement and follow up on actions. Most of the work happens between surveys (Linke 2018, p. 93). Actions

At Ms. Albers' company, the survey runs for three weeks and ends before the Easter holidays. Production is still an open question, because many people there don't have a computer of their own. In one practical case, warehouse staff without a PC workstation were given a room, a kind of polling station (Wiertalla 2013, p. 181). By the way, in QUESTIONSTAR you can use kiosk mode on an iPad for this. That way, employees without a computer of their own can take part too, for example on a tablet in the break room.

Employee survey questions

Clarify the goals first and only turn to standard questions in a second step (Nürnberg 2017, p. 38). A few rules help. The examples are translated from the German sources.

• One topic per question, no extremes. "My supervisor is friendly and competent" is two questions (Borg 2003, cited in Werther 2015, pp. 43-44).
• State the time frame, so "in the past 6 months" instead of "recently" (Thielsch et al. 2012, pp. 231-232).
• Only ask what people can judge. Anyone can answer "Are the goals clear to you?", but hardly "Are the goals the right ones?" (Linke 2018, p. 79).
• Ask about experience, not for a judgment about a person. "My supervisor respects me" is better than "My supervisor is a good leader" (Nürnberg 2017, p. 122).

One scale for all questions

Attitudes are usually measured with a five-point scale (Jacob et al. 2019, p. 187). A textbook shows a symmetric agreement scale from "strongly agree" through "partly agree, partly disagree" to "strongly disagree" (Jacob et al. 2019, p. 188, translated). The midpoint must not be visually highlighted (Jacob et al. 2019, p. 263). A uniform scale saves survey time (Werther 2015, p. 43). In my examples, I use this one:

strongly disagree · somewhat disagree · neither agree nor disagree · somewhat agree · strongly agree

(set apart, only for questions that not everyone can judge) can't judge

"Can't judge" isn't a midpoint

An escape answer such as "don't know" can make sense, but some people use it as a way out (Thielsch et al. 2012, pp. 233-234). You mustn't leave it out entirely, a textbook stresses for knowledge questions, because "don't know" is a substantive answer (Jacob et al. 2019, p. 151). I apply this to questions that someone may lack the knowledge or firsthand experience to answer, for example about processes, other departments or events they didn't witness. On questions about respondents' own attitudes, judgments or wellbeing, everyone has an opinion, so I don't offer it there. Where you offer it, set it apart from the scale.

Let's work out what happens if this answer is counted incorrectly. Suppose 40 employees answer "Problems are raised and resolved promptly here", 8 of them with "can't judge". The other 32 are distributed across points 1 to 5 with 2, 4, 6, 14 and 6 answers. The total is 1 × 2 + 2 × 4 + 3 × 6 + 4 × 14 + 5 × 6 = 114, and the mean is 114 / 32 ≈ 3.56. If you count "can't judge" as 3, you get (114 + 8 × 3) / 40 = 3.45. "I don't know" turns into an opinion, and the value drops. It's better to report the share separately: 20% can't judge this, and that's a result too.

In QUESTIONSTAR, "can't judge" isn't counted in the mean. If you export the data and continue calculating in Excel, SPSS or R, check how the answer is coded there.

16 sample questions by topic

All statements use the scale above. The "Origin" column shows whether a statement comes word for word from a source, is adapted or was worded by me. Statements from German sources are my translations. My own wordings aren't a tested scale.

In addition, there are no more than three open questions:

• "What specifically could improve communication in the company?" (Nürnberg 2017, p. 39, translated)
• "What should we change first so that you can do your work better?" (own)
• "Is there a topic that was missing from this survey?" (own)

Open questions bring ideas, but they take a lot of effort to analyze (Nürnberg 2017, pp. 39-40). Ask for a suggestion instead of asking why. A methods book advises against why questions (Jacob et al. 2019, p. 154). And keep the questionnaire short. One with 100 questions doesn't automatically lead to more change than one with 20 (Werther 2015, p. 48). You'll find more on wording in how to write good survey questions, and further symmetric scales in our list of common Likert scales.

Templates to adapt

If you don't want to start from scratch, you'll find templates in QUESTIONSTAR:

• Employee survey: annual (comprehensive) for the big survey
• Employee pulse quarterly: engagement & enablement and Employee pulse for small teams: informal for pulse surveys
• Employee wellbeing & stress (PSS-10 + WHO-5)
• Mood barometer for a change initiative (ADKAR + eNPS) for change projects
• Hybrid and remote work: setup survey

The wellbeing and stress template is a justified exception to my scale rule. It uses two established instruments with their own frequency scales. The WHO-5 is one of the most widely used questionnaires for subjective psychological wellbeing (Topp et al. 2015, p. 167). The Perceived Stress Scale asks how stressful people find their life situation, and the original version had 14 questions (Cohen et al. 1983, p. 385). Don't convert such instruments to a different scale, and don't use them to assess individuals. If this involves processing health data that can be linked to individuals, you need, in addition to Art. 6, an exception under Art. 9(2) of the GDPR, the EU's General Data Protection Regulation. So coordinate this template especially carefully with the works council and data protection.

For special occasions, there are the Onboarding pulse 30/60/90 days, the Stay survey: proactive retention diagnosis (modular) and the Exit survey: structured departure feedback (modular).

You copy a template into your account with a button inside the opened template. If you don't have an account yet, register for free first. Then check whether the questions fit your goals. You'll find all HR templates under Templates.

Anonymity: what it really means

Wiertalla calls for protecting individual answers and also for not disclosing who took part. Neither taking part nor staying away may lead to disadvantages (Wiertalla 2013, p. 185). With personal links, you therefore distinguish between the anonymity of the answers and the confidentiality of the participation status. Anyone who fears that an honest answer will harm them will rationally choose the socially desirable one (Schnell 2019, p. 44). In one practical case, employees trusted the technology but feared that managers would work backward from percentages (Wiertalla 2013, p. 187).

Anonymous, even with a personal link?

You often want to invite people by email with a personal link and remind only those who haven't answered yet. For that, the system needs an identifier. It can arouse mistrust, which is why dealing with it openly is recommended (Schnell 2019, p. 245). A practitioner's book is stricter: participation is only voluntary if it isn't checked (Nürnberg 2017, p. 68).

By the way, in QUESTIONSTAR you decide for yourself whether to ask for names at all. With the anonymity setting, personal details are stored separately from the answers. At the stricter level, the technical link is cut as well: no identifiers are stored, and the order of the records is randomly shuffled. Answers can then no longer be matched to a person through this technical link. Inferences from answer content, free text or small groups are still possible. You can still send reminders via personal links to everyone who hasn't answered yet. You also decide which users of your account have access to the results or to the personal details at all. It's best to agree with the works council beforehand which level you choose and who gets access.

A textbook advises weighing up whether personalized invitations call the promise of anonymous analysis into question and, if in doubt, doing without them and without tracking responses (Jacob et al. 2019, p. 122). The alternative is a joint thank-you and reminder message to everyone (Jacob et al. 2019, p. 228). In any case, explain your procedure in the invitation.

How large must a group be for its own report?

Before the survey starts, define the level of analysis and the minimum number of responses for a group report (Linke 2018, p. 61). According to Linke, there's no legally binding minimum number, and he advises agreeing on it with the employee representatives (Linke 2018, p. 64). The literature names different thresholds:

For none of these numbers do the sources give an empirical basis. The higher the threshold, the better the protection, but small teams then see only the results of the unit above them (Linke 2018, pp. 62-63). For analyses by demographic characteristics, Linke recommends a higher threshold (Linke 2018, p. 64). For scale reports, what counts is the people who actually responded, not the heads on the org chart (Nürnberg 2017, p. 67). The rule of 50 for comments, on the other hand, means 50 direct and indirect reports, not 50 respondents (Wiertalla 2013, p. 157). My addition: check the minimum number for every question, after deducting skipped answers and, where offered, "can't judge".

Don't confuse this threshold with the precision of the results. As a rule of thumb, subgroups should have at least 20 to 25 cases (Jacob et al. 2019, p. 72). Five answers can be permissible and still hardly interpretable. Tell everyone your threshold before the start (Nürnberg 2017, p. 67).

No threshold protects against working backward

Two reports that differ by only one person reveal that person's answer (Linke 2018, p. 102). Let's work it out. In administration, 46 of 70 respondents agree with the statement "My manager gives me regular feedback on my work", that's ≈ 65.7%. A second report shows "administration, at least one year with the company": 45 of 69, so ≈ 65.2%. Both are far above every threshold in the table. But the difference is exactly one person and one agreement. Anyone who knows that only one new colleague joined in the past year now knows that she agreed. Combining characteristics such as length of service and work group can make individuals identifiable (Linke 2018, p. 64). So only ask for the demographic characteristics you really need (Borg 2003, cited in Wiertalla 2013, p. 155).

In QUESTIONSTAR, you can combine filters in the report, for example department, age group and length of service. That's exactly how small groups arise. QUESTIONSTAR doesn't hide them automatically. You enforce your threshold yourself, through the agreement with the works council, the filters in the report and individual users' access to the data. With many reports, in theory everything would have to be checked against everything else (Linke 2018, p. 103). So it's better to pass on a few reports defined in advance than freely filterable data, and to check group sizes and totals as you do.

Open answers are the biggest gap. With verbatim quotes and in small units, the risk of recognition is particularly high, especially in employee research (Malhotra and Birks 2007, p. 425). Tell employees who reads their comments (Wiertalla 2013, pp. 251-252). And in the workshop afterward, anonymity is partly lifted again (Wiertalla 2013, p. 290). Nobody should be pressured there to disclose their answer.

Works council and data protection

The following overview applies to Germany. I'm not a lawyer and can't tell you what applies legally in your company. I'm only showing which rules usually play a role. Clarify the details with the works council, the data protection officer and, if needed, legal counsel.

Works council. In Germany, the works council (Betriebsrat) is the body elected by the employees to represent them in a company. Its rights are set out in the Works Constitution Act (Betriebsverfassungsgesetz, BetrVG). If there's a works council, talk to it early, ideally during planning. As a rule, it must be informed in good time and comprehensively, and it can request the documents it needs for its tasks (§ 80(2) sentences 1 and 2 BetrVG). Depending on the design, a right of co-determination may be added:

• If the survey asks for personal details, it may need the works council's consent as a personnel questionnaire (Personalfragebogen) (§ 94(1) sentence 1 BetrVG). For a strictly voluntary survey, the Federal Labor Court (Bundesarbeitsgericht, BAG) rejected this in a specific case in 2017 (BAG, decision of November 21, 2017, 1 ABR 47/16). This can't be generalized.
• If the survey runs on an online tool that can link answers or participation to individuals, a right of co-determination over technical equipment may apply (§ 87(1) no. 6 BetrVG).
• If you use personal links for invitations and reminders, you process information about who has already responded, at least temporarily. This participation information is personal data, even if it's only stored under an identifier (Art. 4(1) GDPR; Recital 26 GDPR). Agree with the works council how strict the anonymity setting will be and who has access to results and personal details (§ 87(1) no. 6 BetrVG).

Participation is usually voluntary. The procedure, anonymity rules and the passing on of results are often set down in a works agreement (Betriebsvereinbarung), a written agreement between employer and works council (§ 77(2) sentence 1 BetrVG; Linke 2018, p. 71). The law doesn't specify a fixed minimum group size. In the public sector, the staff council (Personalrat) takes on this role, governed by the Federal Staff Representation Act (Bundespersonalvertretungsgesetz, BPersVG) or the relevant state law (§ 80(1) nos. 15 and 21 BPersVG).

Data protection. The GDPR applies to personal data, and truly anonymous answers aren't covered by it (Recital 26 GDPR; Art. 4(1) GDPR). But genuine anonymity is harder to achieve in online surveys than it sounds, as the calculation above shows. Which legal basis applies to your survey is best clarified by your data protection officer or legal counsel (Art. 6(1) GDPR). If there is a data protection officer, bring them in early. If you collect personal data, inform participants at the time of collection, for example about the purpose, the controller, the legal basis and the storage period (Art. 13 GDPR). Keep it brief but complete, because the privacy notice is often the page where most people drop out (Linke 2018, p. 66). If an external provider processes personal data on your behalf, you usually need a data processing agreement (Art. 28 GDPR).

During planning, Ms. Albers presents the works council with the goal, questionnaire, tool, invitation method, minimum group size, planned reports and schedule. The two sides record this in a works agreement, and the data protection officer checks the privacy notice and the contract with the provider. That's a good start, nothing more.

Pulse survey or annual survey?

The sources describe what a pulse survey is in different ways. A practitioner's book calls it very short and very specific, with about five to ten questions (Nürnberg 2017, p. 37). Linke describes it as a survey of a sample that provides a quick and inexpensive snapshot of the mood (Linke 2018, pp. 33-34).

The annual survey is broad and needs a stable core, otherwise comparisons over time suffer (Linke 2018, pp. 34-35). A pulse is narrow and is suited to tracking progress on priority topics between two large surveys (Nürnberg 2017, p. 100).

I haven't found a study on the best frequency in my sources. Many companies survey annually, as a compromise between impact and cost (Linke 2018, p. 32). A dissertation recommends a two-year cycle to its case company, which surveyed annually, so that there's time for improvements (Spichalsky 2016, pp. 183 and 185). If you ask too often, you risk survey fatigue (Linke 2018, p. 32). And a pulse also needs visible consequences. Werther describes a corporation that turns each of its quarterly pulse surveys into follow-up processes (Werther 2015, p. 57). My advice: only ask as often as you can implement something and report back between two rounds.

Some companies report monthly samples as an average of the last three months (Linke 2018, p. 33). That smooths the curve but reacts with a delay. If agreement drops from 60% to 48% in month 7 and stays at 48%, the rolling value first shows (60 + 60 + 48) / 3 = 56%, and reaches 48% in month 9. So it's better to compare periods that don't overlap, such as quarter with quarter. This conclusion is my own calculation.

Analysis: comparisons, groups, comments

There are three kinds of comparison: between departments, with other companies and with earlier surveys (Nürnberg 2017, p. 82). For comparisons over time, change questions only cautiously (Nürnberg 2017, p. 65). I'd be stricter: same question, same scale. Results above a benchmark aren't the same as consistently good results (Wiertalla 2013, pp. 192-193). External benchmarks can create a false sense of security or discourage people (Linke 2018, p. 106), and it often remains unclear what they're based on (Linke 2018, p. 107). Check the survey method, response rate, population, sample and questionnaire, as with all third-party data (Malhotra and Birks 2007, p. 97).

Worked example: is the difference real?

Ms. Albers looks at the statement "My manager gives me regular feedback on my work". "Somewhat agree" and "strongly agree" count as agreement. Anyone who skipped the statement isn't counted.

Step 1, the response rate. In production, 102 of 150 responded, that's 68%. In administration, 74 of 90, so ≈ 82.2%.

Step 2, the shares. Production: 43 of 96, so ≈ 44.8%. Administration: 46 of 70, so ≈ 65.7%. The gap is ≈ 20.9 percentage points.

Step 3, the uncertainty. It can be illustrated with a 95% confidence interval, calculated here following Wilson (1927). Today this interval is considered more reliable than the simple formula, especially for small groups (Brown et al. 2001, p. 101). For production, it ranges from ≈ 35.2% to ≈ 54.7%, and for administration from ≈ 54.0% to ≈ 75.8%. For the gap, combined from both intervals, it ranges from ≈ 5.6 to ≈ 34.8 points. Zero isn't in it. So the gap is larger than chance makes plausible for groups of this size.

Step 4, the previous year. In 2025, 34 of 88 in production agreed, so ≈ 38.6%. The increase is ≈ 6.2 points, but the interval for it ranges from ≈ −8.0 to ≈ +19.9. An improvement is possible, but it isn't proven.

Step 5, a filter. Ms. Albers filters administration to "under 30" and "less than two years with the company". That leaves 4 answers, and 3 of them agree. "75%" sounds like a finding, but the interval ranges from about 30% to about 95%. Whether the cell reaches your agreed minimum size depends on your rule. With so few answers and such a wide interval, I wouldn't pass it on.

What follows from this? The gap between production and administration is a reason to look more closely, but it doesn't say why. Comparisons alone don't lead to change (Werther 2015, p. 50). And production responded less often. Those who didn't respond may think differently.

For context: total error consists of sampling error, nonresponse error and response error (Malhotra and Birks 2007, pp. 83-84). In a survey of all employees, only the first one disappears. Confidence intervals assume random selection (Jacob et al. 2019, p. 71). So the interval only shows how much a share would fluctuate with so few answers because of random composition. And because the same people often respond in both years, the comparison with the previous year is only an approximation.

How to calculate this with your data

In QUESTIONSTAR, you compare groups in the crosstab, for example agreement by department. I calculated the intervals in the example outside QUESTIONSTAR. To do this, you export the data as an SPSS, CSV or XLSX file and continue in SPSS, R or Python. Or an AI assistant such as ChatGPT or Claude does the calculation via QUESTIONSTAR's MCP interface. Then the assistant does the calculation, not QUESTIONSTAR. Questions marked as personal data are not transferred to the assistant when the protection setting is on. Unmarked answers, especially free text, can still contain personal information. Before transferring, check which answers go to the AI provider and whether its contract terms suit your data. My tip: mark open questions where someone might write something personal as personal data.

Reading comments

Open comments are written mainly by dissatisfied people. In one study, their comments tended to be negative, and the more dissatisfied someone was, the longer they were (Poncheri et al. 2008, cited in Schnell 2019, p. 71). The data come from a military organization but match Linke's experience (Linke 2018, p. 40). So how often a topic comes up in comments says little about how widespread a problem is. Still, comments help with action planning, because they're more concrete than scale values (Linke 2018, p. 40).

Categorizing them is time-consuming and error-prone (Schnell 2019, p. 72). With automatic coding, the meaning of a comment can even change (Linke 2018, p. 40). These sources predate today's language models. So read an AI summary as a suggestion and check samples against the originals. And sort comments by urgency first, not by frequency. That's my suggestion, not a finding from the literature. A single note about an occupational safety hazard can matter more than twenty complaints about the cafeteria. One such comment, about worn saw blades, appears in a list of comments that Nürnberg reprints (Nürnberg 2017, pp. 131-132). Ask what consequences a topic has and how well the scale values support it. Without a fixed procedure, comments are either overlooked or overemphasized (Wiertalla 2013, pp. 157 and 252).

From results to actions

The biggest risk of an employee survey is that, from the employees' point of view, nothing at all happens afterward (Linke 2018, p. 21). That's a practitioner's judgment, not a measurement. People who see that nothing happens often don't take part in the next survey (Linke 2018, p. 21). Experts even advise not running a survey at all rather than running a poor one (Bungard et al. 1997, cited in Wiertalla 2013, p. 108). "Nothing happens" can also mean that actions are underway, but nobody notices them, or nobody connects them with the survey (Nürnberg 2017, p. 95).

An action plan in seven columns

In the assembly team's feedback workshop, the team chooses two topics: information gets lost at shift handover, and suitable tools are missing. In one documented company format, each team chose no more than three topics (Wiertalla 2013, p. 206). That's a company rule, not an ideal number. A few important topics are better than many small actions (Nürnberg 2017, p. 94).

Before the action comes the question of causes and of who is allowed to decide (Linke 2018, p. 90). If several teams hit the same limit, a local topic becomes a central one (Linke 2018, p. 90). Without follow-up, day-to-day business crowds out implementation (Linke 2018, p. 83). And "deliberately no action" is a valid entry if there's no need for change (Borg 2002, cited in Wiertalla 2013, p. 78). In that case, write down the reason.

Ask again in the next wave

Some companies ask in the next survey whether employees were informed about results and actions and involved in them (Linke 2018, pp. 126-127). For this, Ms. Albers adds two statements with the same scale: "I was informed clearly about the actions taken after the last survey." and "I was able to take part in developing actions in my team." These questions measure how the follow-up process is experienced. They don't prove that an action worked. Cross-sectional data don't prove cause-and-effect relationships (Spichalsky 2016, p. 169), and other events can shift the answers even during the field period (Spichalsky 2016, p. 171).

Careful with target values and bonuses

As an example of a measurable goal, Linke names an increase of at least 0.5 scale points in the next survey (Linke 2018, p. 113). Let's work it out, for two independent waves and an assumed standard deviation of one point. With 8 answers per wave, the difference between two team means fluctuates by ≈ ±0.98 points through chance alone (95% range in the normal approximation, 1.96 × √2 × 1.0 / √8). With about 31 or more responses per wave, an increase of 0.5 points lies just outside this range. If the same people answer in both waves, the calculation looks different.

This is even more true of bonuses. Anyone who links managers' career progression to the results creates incentives to influence them through pressure or cooperation (Linke 2018, p. 19). A dissertation suggests instead including the implementation of actions in target agreements (Spichalsky 2016, p. 181).

Frequently asked questions about employee surveys

How often should you run an employee survey?

There's no proven best frequency. Many companies survey annually (Linke 2018, p. 32). My advice: as often as you can implement something and report back between two rounds. In between, short pulse surveys can track progress.

Is an employee survey anonymous?

Answers shouldn't be traceable to individual employees. With personal links, the system can still process the participation status. Be open about who can see this status. Even without a technical link, small groups or the content of answers can make people identifiable. You set the threshold for small groups in advance with the works council and enforce it yourself. More under Anonymity.

Does the works council have to agree?

In Germany, that depends on the design. As a rule, the works council must be informed in good time and comprehensively (§ 80(2) BetrVG). If the survey is a personnel questionnaire within the meaning of § 94(1) BetrVG, it needs the works council's consent. For a strictly voluntary employee survey, the Federal Labor Court rejected this classification in one case (BAG, decision of November 21, 2017, 1 ABR 47/16, para. 31). If an online tool can link answers or participation to individuals, a right of co-determination may apply (§ 87(1) no. 6 BetrVG). The rules are often set down in a works agreement. Talk to the works council or staff council early.

Which questions belong in an employee survey?

Questions on topics you can and want to change, such as leadership, collaboration, information, working conditions and development. Start with an overall question on satisfaction, and use a symmetric scale with a neutral midpoint. Offer "can't judge" only for questions that not everyone can answer from their own experience, and set it apart from the scale. You'll find 16 examples under Questions.

What response rate should you aim for?

I can't give you a fixed number. Treating the response rate on its own as a measure of quality is misleading (Schnell 2019, p. 159). What matters is whether respondents and nonrespondents differ. The bias is the share of nonrespondents times this difference (Schnell 2019, p. 160).

Two cases. Case 1: 80% respond. Of them, 70% agree, but of the missing 20%, only 30% do. The true value is 0.8 × 70 + 0.2 × 30 = 62%. The survey shows 70%, 8 percentage points too high. Case 2: only 50% respond, but both groups think the same, 62% each. Then the survey is exactly right. In a meta-analysis, the nonresponse rate explained only 11% of the variance in nonresponse bias (Groves and Peytcheva 2008, cited in Schnell 2019, pp. 160-161). So the question of the response rate at which a survey becomes representative can't be answered this way (Schnell 2019, p. 163).

It's still worth reaching many people. But don't chase the rate, because a high rate can also be the result of social pressure (Wiertalla 2013, pp. 183-184). Unusable answers are deducted from the responses (Nürnberg 2017, p. 74). That's why I report the response rate and the valid answers per question separately.

What is a pulse survey?

A short survey with about five to ten questions on one topic (Nürnberg 2017, p. 37) that runs at shorter intervals than the big survey (Werther 2015, p. 29). More under Pulse survey.

Sources

Borg, I. (2002). Mitarbeiterbefragungen - kompakt [Employee surveys: compact]. Göttingen: Hogrefe. In German. Cited in Wiertalla (2013), p. 78.

Borg, I. (2003). Führungsinstrument Mitarbeiterbefragung [The employee survey as a management tool]. Göttingen: Hogrefe. In German. Cited in Werther (2015), pp. 27-28, 30 and 43-44, Linke (2018), pp. 75-76, and Wiertalla (2013), p. 155.

Brown, L. D., Cai, T. T. and DasGupta, A. (2001). Interval Estimation for a Binomial Proportion. Statistical Science, 16(2), 101-133. DOI 10.1214/ss/1009213286.

Bungard, W., Jöns, I. and Schultz-Gambard, J. (1997). Sünden bei der Mitarbeiterbefragung. Zusammenfassung der wichtigsten Fehler und Fallgruben [Sins in employee surveys: a summary of the main mistakes and pitfalls]. In W. Bungard and I. Jöns (eds.), Mitarbeiterbefragung. Ein Instrument des Innovations- und Qualitätsmanagements [The employee survey: an instrument of innovation and quality management] (pp. 441-455). Weinheim: Beltz, Psychologie Verlags Union. In German. Cited in Wiertalla (2013), p. 108.

Bungard, W., Müller, K. and Niethammer, C. (2007). Mitarbeiterbefragung - was dann ...? MAB und Folgeprozesse erfolgreich gestalten [Employee surveys, and then what? Designing surveys and follow-up processes successfully]. Heidelberg: Springer. In German. Cited in Werther (2015), pp. 27-28.

Cohen, S., Kamarck, T. and Mermelstein, R. (1983). A global measure of perceived stress. Journal of Health and Social Behavior, 24(4), 385-396. DOI 10.2307/2136404.

Gallup (2026). Engagement Index Deutschland 2025 [Engagement Index Germany 2025]. In German. Online, accessed November 9, 2026.

Groves, R. and Peytcheva, E. (2008). The Impact of Nonresponse Rates on Nonresponse Bias: A Meta-Analysis. Public Opinion Quarterly, 72(2), 167-189. Cited in Schnell (2019), pp. 160-161.

Jacob, R., Heinz, A. and Décieux, J. P. (2019). Umfrage. Einführung in die Methoden der Umfrageforschung [Surveys: an introduction to survey research methods] (4th ed.). Berlin/Boston: De Gruyter Oldenbourg. In German.

Levenson, A. (2014). Employee Surveys That Work. Berrett-Koehler. Cited in Nürnberg (2017), p. 26.

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.

Locke, E. (1976). The nature and causes of job satisfaction. In M. Dunnette (ed.), Handbook of industrial and organizational psychology (pp. 1297-1343). Chicago: Rand McNally. Cited in Spichalsky (2016), p. 38.

Malhotra, N. K. and Birks, D. F. (2007). Marketing Research. An Applied Approach (3rd European ed.). Harlow: Pearson Education.

Nürnberg, V. (2017). Mitarbeiterbefragungen. Ein effektives Instrument der Mitbestimmung [Employee surveys: an effective instrument of co-determination] (1st ed.). Freiburg: Haufe-Lexware. In German.

Poncheri, R. M., Lindberg, J. T., Thompson, L. F. and Surface, E. A. (2008). A Comment on Employee Surveys: Negativity Bias in Open-Ended Responses. Organizational Research Methods, 11(3), 614-630. DOI 10.1177/1094428106295504. Cited in Schnell (2019), p. 71.

Schaufeli, W. B., Salanova, M., González-Romá, V. and Bakker, A. B. (2002). The Measurement of Engagement and Burnout: A Two Sample Confirmatory Factor Analytic Approach. Journal of Happiness Studies, 3, 71-92. DOI 10.1023/A:1015630930326.

Schnell, R. (2019). Survey-Interviews. Methoden standardisierter Befragungen [Survey interviews: methods of standardized questioning] (2nd ed.). Wiesbaden: Springer VS. In German. DOI 10.1007/978-3-531-19901-6.

Spichalsky, K. (2016). Change Management und Mitarbeiterbefragungen. Konzeptionelle Überlegungen und empirische Erkenntnisse [Change management and employee surveys: conceptual considerations and empirical findings]. Wiesbaden: Springer. In German. DOI 10.1007/978-3-658-14096-0.

Thielsch, M. T., Lenzner, T. and Melles, T. (2012). Wie gestalte ich gute Items und Interviewfragen? [How do I design good items and interview questions?] In M. T. Thielsch and T. Brandenburg (eds.), Praxis der Wirtschaftspsychologie. Bd. II, Themen und Fallbeispiele für Studium und Anwendung [Business psychology in practice, vol. II: topics and case studies for study and application] (pp. 221-240). Münster: Monsenstein und Vannerdat. In German.

Topp, C. W., Østergaard, S. D., Søndergaard, S. and Bech, P. (2015). The WHO-5 Well-Being Index: A Systematic Review of the Literature. Psychotherapy and Psychosomatics, 84(3), 167-176. DOI 10.1159/000376585.

Werther, S. (2015). Einführung in Feedbackinstrumente in Organisationen. Vom 360°-Feedback bis hin zur Mitarbeiterbefragung [An introduction to feedback instruments in organizations: from 360° feedback to the employee survey]. Wiesbaden: Springer (essentials). In German. DOI 10.1007/978-3-658-10497-9.

Wiertalla, K. (2013). Der Einsatz von Mitarbeiterbefragungen als Lerninstrument in Unternehmen. Dargestellt anhand des Praxisbeispiels der Mitarbeiterbefragung bei der Ingram Micro Distribution GmbH [Employee surveys as a learning instrument in companies, illustrated by the employee survey at Ingram Micro Distribution GmbH]. Doctoral dissertation, University of Siegen. In German.

Wilson, E. B. (1927). Probable Inference, the Law of Succession, and Statistical Inference. Journal of the American Statistical Association, 22(158), 209-212. DOI 10.1080/01621459.1927.10502953.

Legal sources

• Works Constitution Act (Betriebsverfassungsgesetz, BetrVG), § 77, § 80, § 87, § 94. In German.
• Federal Staff Representation Act (Bundespersonalvertretungsgesetz, BPersVG), § 80. In German.
• General Data Protection Regulation (GDPR), Regulation (EU) 2016/679, Art. 4, 6, 9, 13, 28 and Recital 26, EUR-Lex
• Federal Labor Court (Bundesarbeitsgericht), decision of November 21, 2017, 1 ABR 47/16, decision. In German.

You'll find more on using QUESTIONSTAR in HR under HR surveys.

I wish you a survey after which your colleagues say: it was worth answering honestly.

Related articles:

• 360-degree feedback: process, sample questions and common mistakes
• How to write good survey questions
• Common Likert scales: examples with labels

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.