Collecting survey responses is often the easy part. The difficulty starts when you open the data file: incomplete responses, text where numbers should be, reverse-worded items, a Likert scale you are not sure whether to treat as ordinal or interval, and a supervisor asking which test you used and why.
Survey data analysis is a sequence of decisions, and each one needs to be defensible in your methods and results chapters.
This guide walks through that sequence for a typical dissertation, thesis or research paper: preparing and cleaning the data, scoring scales, checking reliability, describing the sample, choosing statistical tests that match your research questions, analyzing open-ended responses and reporting the results.
It uses examples from common survey designs and explains where people most often go wrong.
The survey analysis sequence
To analyze survey data, first clean it (remove invalid and duplicate responses, handle missing data, recode reverse-worded items), then compute scale scores and check reliability, usually with Cronbach's alpha. Describe the sample with frequencies, percentages, means and standard deviations. Then choose tests that match each research question and variable type: chi-square for two categorical variables, t-tests or ANOVA (or their non-parametric equivalents) for group differences, correlation and regression for relationships. Report results with test statistics, significance and effect sizes.
The Survey Analysis Workflow
| Step | What you do | What you report |
|---|---|---|
| 1. Prepare | Export data, create a codebook, label variables and values | Software used |
| 2. Clean | Screen for incomplete, duplicate, speeding or straight-lined responses | Number of responses removed and why |
| 3. Missing data | Assess amount and pattern; choose a handling method | Method used and its justification |
| 4. Recode and score | Reverse-code items, compute scale totals or means | Scoring procedure |
| 5. Reliability | Cronbach's alpha (or omega) for each multi-item scale | Reliability coefficients |
| 6. Describe | Sample characteristics and scale descriptives | Demographics table, means and SDs |
| 7. Test | Inferential tests for each research question or hypothesis | Test statistics, p-values, effect sizes |
| 8. Interpret | Relate results to the questions and literature | Discussion chapter |
Cleaning Survey Data
Data cleaning decides which responses count, so it must be done systematically and reported honestly. Common checks include incomplete responses (for example, participants who stopped after the consent page), duplicate submissions, respondents who failed attention-check items, implausibly fast completion times, and straight-lining, where someone selects the same option for every item in a grid.
Decide your rules before you look at the results, apply them consistently, and record how many responses each rule removed.
Reporting example: "Of 318 responses received, 41 were excluded: 29 did not complete beyond the demographic section, 7 failed both attention checks and 5 were duplicates identified by matching timestamps and responses. The final sample comprised 277 participants."
Handling Missing Data
Almost every survey has some missing answers. First look at how much is missing and whether it follows a pattern (for instance, many people skipping a sensitive income question).
Options include listwise deletion (dropping any case with missing values, simple but it reduces the sample), pairwise deletion, calculating scale means from the items answered when only a small number are missing, and multiple imputation for more substantial or patterned missingness.
Whatever you choose, state it and justify it. Simply ignoring missing data without saying so is one of the most common criticisms in dissertation feedback.
Coding and Scoring Likert Scales
Likert-type items (for example, 1 = strongly disagree to 5 = strongly agree) are the backbone of most surveys. Before scoring, reverse-code negatively worded items so that a high score means the same thing across all items.
On a 1 to 5 scale, the reversed score is 6 minus the original. Then compute a total or mean score for each scale, following the scoring instructions published with the instrument if you used a validated one.
| Question | Common approach |
|---|---|
| Single Likert item | Usually treated as ordinal: report frequencies, median, mode; compare with non-parametric tests |
| Multi-item Likert scale (summed or averaged) | Often treated as approximately interval: report mean and SD; parametric tests if assumptions are reasonable |
| Agreement summary | Collapsing "agree" and "strongly agree" can help presentation but loses detail; report the full distribution too |
The ordinal versus interval question is debated in methods literature. What examiners want is that you make a reasoned choice, cite support for it, and apply it consistently.
Checking Reliability and Validity
For each multi-item scale, report internal consistency. Cronbach's alpha is the most widely used measure; values around .70 or higher are conventionally treated as acceptable for research purposes, with higher thresholds sometimes expected for high-stakes measures. Very high values (above about .95) can indicate redundant items.
If alpha is low, check the item-total statistics to see whether a particular item, often a reverse-worded one that was not recoded, is pulling it down. If you developed new items or adapted a scale substantially, your supervisor may also expect exploratory or confirmatory factor analysis to examine the scale's structure.
Choosing the Right Statistical Test
The right test depends on your research question and the type of variables involved. This table covers the most common survey scenarios.
| Research question type | Variables | Parametric test | Non-parametric alternative |
|---|---|---|---|
| Are two categorical variables related? | Categorical x categorical | Chi-square test of independence | Fisher's exact test (small expected counts) |
| Do two groups differ? | Two groups, continuous outcome | Independent samples t-test | Mann-Whitney U |
| Do three or more groups differ? | Several groups, continuous outcome | One-way ANOVA with post hoc tests | Kruskal-Wallis |
| Did scores change (pre and post)? | Same respondents measured twice | Paired samples t-test | Wilcoxon signed-rank |
| Are two continuous variables related? | Continuous x continuous | Pearson correlation | Spearman correlation |
| Which factors predict an outcome? | Several predictors, continuous outcome | Multiple linear regression | (Consider transformations or robust methods) |
| Which factors predict a yes/no outcome? | Several predictors, binary outcome | Binary logistic regression | Not applicable |
Check the assumptions of each parametric test (approximate normality of residuals, homogeneity of variance, independence) and report what you checked.
With very large samples, small departures from normality matter less; with small samples, consider the non-parametric option. Regression has its own set of assumptions and diagnostics, covered in our guide to regression analysis help.
Worked Example: Answering a Research Question
Research question: Does job satisfaction differ between full-time and part-time retail employees?
Data: Job satisfaction measured with a 6-item scale (alpha = .84), averaged to a 1 to 5 score; employment status coded full-time or part-time.
Analysis: Independent samples t-test after checking group distributions and Levene's test for equality of variances.
Reporting (illustrative format): "Full-time employees (M = 3.62, SD = 0.71) reported higher job satisfaction than part-time employees (M = 3.28, SD = 0.80), t(275) = 3.67, p < .001, d = 0.45."
The numbers here illustrate the APA-style format only; your results must come from your own data. Note the elements: group means and standard deviations, the test statistic with degrees of freedom, the p-value and an effect size.
Analyzing Open-Ended Responses
Open-ended questions add depth but are often left unanalysed. Treat them as qualitative data: read all responses, develop codes, group codes into categories or themes, and count how often each appears if that is useful.
Short, numerous responses suit content analysis with a coding frame; longer, reflective answers suit thematic analysis. Report the coding approach briefly and use short quotations to illustrate each theme. If two people code the data, reporting agreement strengthens the analysis.
Sample Size, Response Rate and Bias
Your analysis can only be as convincing as the sample behind it. Report how participants were recruited, how many were invited (if known), how many responded and how many remained after cleaning.
A response rate is only meaningful when you know the number invited; with open online links you often cannot calculate one, and you should say so rather than guess. Compare your sample's characteristics with the population you are interested in where data exist, for instance the gender or age profile of a profession, and note any obvious differences.
Sample size also affects which analyzes are sensible. Subgroup comparisons with only a handful of respondents per group are unstable, chi-square tests need adequate expected counts in each cell, and regression models need enough cases for the number of predictors.
If your sample is smaller than planned, simplify the analysis rather than forcing a complex model, and acknowledge the reduced power in your limitations.
Software Options
| Software | Strengths |
|---|---|
| SPSS | Menu-driven, widely used in social sciences, health and education |
| R | Free, powerful, reproducible scripts; steeper learning curve |
| Stata | Common in economics and public health; strong for survey weighting |
| Jamovi or JASP | Free, menu-driven, APA-style output |
| Excel | Fine for cleaning and descriptives; limited for inferential tests |
| NVivo or similar | Coding open-ended responses |
Reporting Survey Results
- Start with the response rate (if known), exclusions and final sample size.
- Present a demographics table, then scale descriptives and reliability.
- Organize inferential results by research question or hypothesis, in order.
- Report test statistics, degrees of freedom, p-values and effect sizes.
- Use tables for many results; describe the key ones in text.
- Do not interpret in the results chapter beyond stating what was found; save meaning for the discussion.
- Do not describe non-significant results as "trends" or "almost significant".
Common Mistakes
| Mistake | Consequence |
|---|---|
| Forgetting to reverse-code items | Low reliability and misleading scale scores |
| Running many tests without a plan | Inflated chance of false positives |
| Treating correlation as causation | Overclaiming in the discussion |
| Reporting only p-values | Reader cannot judge the size of effects |
| Ignoring a low response rate | Unacknowledged non-response bias |
| Generalizing from a convenience sample | Conclusions the design cannot support |
How GradeEssays Helps with Survey Data Analysis
Send your data file (SPSS, Excel, CSV, R or Stata), your questionnaire, your research questions or hypotheses and any methods chapter or supervisor feedback. An analyst cleans and checks the data, scores and tests the reliability of your scales, runs the appropriate descriptive and inferential analyses, and explains the output.
You can also receive APA-style tables and a model results write-up to learn from. We analyze only the data you collected and never alter or invent responses. Revisions within the scope of your original request are free, and your identity is never shared with the analyst.
Turn Your Survey into Clear Findings
Share your data file, questionnaire and the questions you need answered. An analyst checks, analyses and helps you report the results correctly.
Start My Survey AnalysisFree revisions · Full refund if late · Written from scratch for your order
Frequently Asked Questions
Individual Likert items are ordinal. Summed or averaged multi-item scales are often treated as approximately interval, which allows parametric tests. Make a reasoned choice, support it and apply it consistently.
Around .70 or above is conventionally treated as acceptable for research scales, though acceptable values depend on the purpose and number of items. Very high values can indicate redundant items.
Set exclusion rules in advance, apply them consistently and report how many responses were removed. For scattered missing items, choose and report a method such as pairwise deletion or multiple imputation.
It depends on your analysis. Power analysis, often done with free tools such as G*Power, estimates the sample needed to detect an effect of a given size. Descriptive surveys aiming to estimate population percentages use different calculations.
On a multi-item scale score, a t-test is commonly used if assumptions are reasonable. For single items, a Mann-Whitney U test is usually the safer choice.
Code the responses, group codes into categories or themes, and report them with short quotations, using content analysis or thematic analysis depending on response length.
Weighting adjusts a sample to match known population characteristics. It is common in large national surveys; most student surveys with convenience samples do not weight, but they should acknowledge the sampling limitation.
Sample and exclusions, demographics, scale descriptives and reliability, then results organized by research question with test statistics and effect sizes.