Design a Survey and Analyze the Data

Overview

Surveys are one of the most powerful tools in science, business, and public policy — and one of the most misused. In this project you'll design a short survey around a real question you're curious about, collect at least 20 genuine responses, organize and clean the data, calculate statistics, and create charts that honestly represent what you found.

The emphasis is on doing it right: asking questions that don't lead the respondent, identifying bias in your sample, and choosing the right type of chart for each type of data.

What You'll Learn

• The difference between categorical and numerical data and how to summarize each
• How to design unbiased survey questions and identify sampling bias
• When to use bar charts, histograms, and pie charts
• How to calculate mean, median, mode, and range from a real dataset

Background

Survey data comes in two types. Categorical data puts responses into groups: "What is your favorite season?" has four possible categories. Numerical data measures quantities: "How many hours of sleep did you get last night?" can be any number. Each type requires different statistical tools and different chart types.

For categorical data, the most useful summaries are frequencies (how many in each category) and proportions (what fraction). Bar charts and pie charts show this. For numerical data, you want measures of center (mean, median, mode) and spread (range, standard deviation). Histograms and box plots show this.

Sampling bias is the biggest threat to survey validity. If you only ask your friends, your sample is not representative of the broader population. If your question is "Shouldn't people eat healthier?" you've loaded the question toward a yes answer. Good survey design minimizes both sampling bias (who you ask) and question bias (how you ask).

Materials

Everything on this list is either free or available at a dollar store. No lab equipment required.

• Paper and pencil for question design Already own
• Google Forms (free at forms.google.com) Free
• Google Sheets or Excel for analysis Free / Already own
• At least 20 willing survey respondents Free

Step-by-Step Procedure

1. 1

   Choose a question and design the survey

   Start with a genuine question you want to answer: Do people in your community sleep enough? Do students prefer project-based or test-based learning? Does screen time affect self-reported happiness? Write your research question at the top of a planning sheet.
   
   Design 5–8 survey questions that help answer it. Include a mix of types: some multiple choice (categorical), some scale questions ("Rate 1–5"), some short numerical ("How many hours per night?"), and one open-ended. Write each question, then check it: Is it leading? Is it double-barreled (asking two things at once)? Is it clear to someone who has no context?

   💡

   Avoid double negatives and "always/never" language. "Do you never eat breakfast?" is confusing. "How often do you eat breakfast?" with response options is much cleaner.

2. 2

   Build and distribute the survey

   Create your survey in Google Forms — it's free and automatically collects responses in a spreadsheet. Add your questions in the correct format (multiple choice, short answer, linear scale). Preview the form from a respondent's perspective and answer it yourself to check for confusing wording.
   
   Share the link with at least 20 people. Try to reach people outside your immediate friend group to reduce sampling bias. Note who you asked and who responded — this "response rate" is important context for your results.

   💡

   Add a brief introduction at the top of the form: what the survey is about, who you are, and that responses are anonymous. This increases honest responses.

3. 3

   Clean and organize the data

   Open the auto-generated Google Sheet of responses. Look for: incomplete responses (rows with many blank cells), nonsensical answers (age = 999, sleep hours = 47), and inconsistent formatting (some wrote "yes," others "Yes," others "YES"). Document any cleaning decisions you make and why — data cleaning is a real, important step in every statistical analysis.
   
   Create a summary table: for each question, list the number of valid responses and any data quality notes.

4. 4

   Calculate statistics and create charts

   For each numerical question: calculate the mean (sum ÷ count), median (middle value when sorted), mode (most common value), and range (max − min). For each categorical question: calculate the count and percentage in each category. Enter all of these in a clean summary table.
   
   Create at least three charts: a bar chart for a categorical question, a histogram for a numerical question, and one more of your choice. Use clear titles, labeled axes, and a legend if needed. For each chart, write one sentence interpreting what it shows.

   💡

   The mean is sensitive to outliers; the median is not. If someone answered "10 hours of sleep" when everyone else answered 6–8, the mean shifts but the median barely moves. Report both when relevant.

5. 5

   Write your findings and identify limitations

   Write a one-page summary of what you found. State your research question, describe your sample (who answered, how many, any likely biases), present your key findings using specific numbers from your analysis, and state your conclusions clearly. Then write a "Limitations" section: What can't you conclude from this data? Who is missing from your sample? What questions would a follow-up study need to answer?
   
   This limitation section is what separates good statistics from misleading statistics. Every honest study includes it.

Expected Results

Your finished project will include: a written survey with 5–8 questions, a spreadsheet of at least 20 responses, a cleaned dataset with any errors or outliers noted, summary statistics for at least two numerical questions, and at least three charts that accurately represent your findings. You'll also include a section identifying the limitations of your data.

The most valuable output isn't necessarily a surprising result — it's a set of conclusions you can actually defend based on your data, with honest acknowledgment of what your data can and cannot show.

Going Further