Survey Design

Survey Design — Survey & Instrument Design Agent

A comprehensive survey and instrument design tool for faculty and researchers. Covers the full lifecycle of survey-based research: from construct definition and item writing through pilot testing, validation, deployment, and analysis of survey data.

Overview

Survey research is one of the most widely used methods across social sciences, health sciences, education, and business. Despite its apparent simplicity, designing a valid and reliable survey instrument requires systematic attention to construct definition, item wording, response format, sampling, bias mitigation, and psychometric validation.

This skill treats survey design as a scientific process, not an art. Every design decision should be justified and documented.

When to Use This Skill

This skill should be used when:

• Designing a new survey or questionnaire from scratch
• Adapting an existing instrument for a new population or context
• Writing Likert-scale items or other structured response formats
• Developing interview protocols or focus group guides
• Planning sampling strategies for survey research
• Conducting pilot tests and cognitive interviews
• Validating instruments (reliability and validity analysis)
• Selecting online survey platforms (Qualtrics, REDCap, Google Forms)
• Improving response rates and reducing bias
• Conducting cultural adaptation and translation of instruments
• Teaching research methods courses that include survey design

Does NOT Trigger

Core Capabilities

1. Survey Design Process

The 10-Step Survey Design Framework:

Step 1:  Define research objectives and constructs
         What do you want to measure? What are your research questions?
              │
Step 2:  Review existing instruments
         Has someone already validated an instrument for your construct?
              │
Step 3:  Define the target population and sampling frame
         Who will you survey? How will you reach them?
              │
Step 4:  Choose survey mode
         Online, paper, phone, in-person, mixed-mode?
              │
Step 5:  Write items and response options
         Craft questions that are clear, unambiguous, and aligned to constructs
              │
Step 6:  Design survey structure and flow
         Organize sections, add skip logic, manage survey length
              │
Step 7:  Expert review
         Subject matter experts and methodologists evaluate the instrument
              │
Step 8:  Cognitive interviews and pilot testing
         Test with a small sample from the target population
              │
Step 9:  Psychometric validation
         Reliability analysis, factor analysis, validity assessment
              │
Step 10: Deploy, monitor, and analyze
         Launch survey, track response rates, clean and analyze data

2. Construct Definition and Operationalization

Before writing a single item, define what you are measuring.

Construct Mapping Template:

## Construct Map

### Construct: [Name]
### Definition: [Precise conceptual definition with citation]

### Dimensions/Facets:
1. [Dimension 1] — [Definition]
   - Indicators: [Observable behaviors or attitudes]
   - Example items: [Draft items]

2. [Dimension 2] — [Definition]
   - Indicators: [Observable behaviors or attitudes]
   - Example items: [Draft items]

3. [Dimension 3] — [Definition]
   - Indicators: [Observable behaviors or attitudes]
   - Example items: [Draft items]

### Related but Distinct Constructs:
- [Construct A] — How it differs from your construct
- [Construct B] — How it differs from your construct

### Nomological Network:
- Should correlate positively with: [Constructs]
- Should correlate negatively with: [Constructs]
- Should be unrelated to: [Constructs]

3. Item Writing

Question Types and When to Use Them

Likert Scale Design

Number of Points:

Likert Scale Labeling:

FULLY LABELED (recommended for clarity):
Strongly Disagree | Disagree | Neutral | Agree | Strongly Agree

END-ANCHORED ONLY (acceptable for experienced respondents):
Strongly Disagree | 2 | 3 | 4 | Strongly Agree

AGREEMENT:        Strongly Disagree → Strongly Agree
FREQUENCY:        Never → Always
IMPORTANCE:       Not at all Important → Extremely Important
SATISFACTION:     Very Dissatisfied → Very Satisfied
LIKELIHOOD:       Very Unlikely → Very Likely
QUALITY:          Very Poor → Excellent

Item Writing Rules

DO:

1. Use simple, clear language (avoid jargon, acronyms, technical terms unless your population uses them)
2. Ask about one thing per item (no double-barreled questions)
3. Use specific time frames ("In the past 30 days..." not "Do you ever...")
4. Match the response scale to the question stem
5. Include both positively and negatively worded items (with caution — see pitfalls)
6. Pilot test items with your target population
7. Write 2-3x more items than you need (expect to cut during validation)

DO NOT:

1. Use leading or loaded language ("Don't you agree that...")
2. Use double negatives ("How much do you disagree with not implementing...")
3. Assume knowledge ("Rate the effectiveness of the Delphi method" — respondent may not know it)
4. Use absolutes ("always," "never," "all," "none") unless measuring frequency
5. Create unnecessarily long items (aim for under 20 words per item)
6. Use hypothetical scenarios when asking about actual behavior

Examples of Item Revisions:

POOR: "How satisfied are you with the quality and timeliness of feedback?"
       (Double-barreled: quality AND timeliness)
FIX:  Item 1: "How satisfied are you with the quality of feedback you receive?"
      Item 2: "How satisfied are you with the timeliness of feedback you receive?"

POOR: "Students should not be required to not attend classes."
       (Double negative)
FIX:  "Class attendance should be mandatory."

POOR: "Do you agree that the new policy is beneficial?"
       (Leading — assumes the policy is beneficial)
FIX:  "The new policy has been beneficial to my work."
       (Neutral stem; let the Likert scale capture agreement/disagreement)

POOR: "Rate your teaching effectiveness." (1-5)
       (Socially desirable response; no reference frame)
FIX:  "In the past semester, how often did you use student feedback
       to modify your teaching?" (Never / Rarely / Sometimes / Often / Always)

4. Survey Structure and Flow

Recommended Survey Organization:

## Survey Structure Template

### Page 1: Welcome and Consent
- Study title, purpose, estimated time
- Consent checkbox (mandatory before proceeding)
- Contact information for questions

### Page 2: Screening Questions (if applicable)
- Eligibility criteria
- Route ineligible respondents to end-of-survey message

### Page 3-N: Main Content Sections
- Group by topic/construct
- Progress bar visible
- Section headers with brief context
- Start with engaging, easy questions
- Place sensitive questions in the middle (after rapport, before fatigue)
- Use skip logic to hide irrelevant questions

### Page N+1: Demographics
- Place at the END (reduces dropout from sensitive questions early)
- Include only demographics you will actually analyze
- Provide "Prefer not to answer" option for sensitive items

### Final Page: Thank You
- Thank participant
- Provide debriefing information
- Share contact info for results
- Remind of withdrawal procedure

Skip Logic Design:

Q1: Do you supervise graduate students?
    ( ) Yes → Show Q2-Q5 (supervision questions)
    ( ) No  → Skip to Q6

Q3: How many students do you currently supervise?
    [Number input]
    If Q3 > 5 → Show Q4 (workload management question)
    If Q3 ≤ 5 → Skip to Q5

Q10: Would you like to participate in a follow-up interview?
     ( ) Yes → Show Q11 (contact information)
     ( ) No  → Skip to end

5. Sampling Strategies

Probability Sampling (every member of population has a known, non-zero chance of selection):

Non-Probability Sampling (no guarantee of representativeness):

Sample Size Determination:

For descriptive surveys (estimating proportions):
n = (Z² × p × (1-p)) / E²

Where:
  Z = Z-score for confidence level (1.96 for 95%)
  p = Expected proportion (use 0.5 if unknown — most conservative)
  E = Margin of error (e.g., 0.05 for ±5%)

Example: 95% confidence, 5% margin of error, unknown proportion
n = (1.96² × 0.5 × 0.5) / 0.05² = 384.16 → 385 respondents

Adjust for finite population:
n_adj = n / (1 + (n-1)/N)
Where N = population size

Adjust for expected response rate:
n_needed = n_adj / expected_response_rate
Example: 385 / 0.30 = 1,284 invitations needed for 30% response rate

For comparative surveys (detecting differences between groups):

# Power analysis for two-group comparison
from scipy import stats
import numpy as np

def sample_size_two_groups(effect_size, alpha=0.05, power=0.80):
    """
    Calculate sample size per group for independent samples t-test.

    effect_size: Cohen's d (0.2=small, 0.5=medium, 0.8=large)
    alpha: significance level
    power: desired statistical power
    """
    z_alpha = stats.norm.ppf(1 - alpha/2)
    z_beta = stats.norm.ppf(power)
    n = 2 * ((z_alpha + z_beta) / effect_size) ** 2
    return int(np.ceil(n))

# Examples
print(f"Small effect (d=0.2):  {sample_size_two_groups(0.2)} per group")   # 394
print(f"Medium effect (d=0.5): {sample_size_two_groups(0.5)} per group")   # 64
print(f"Large effect (d=0.8):  {sample_size_two_groups(0.8)} per group")   # 26

6. Response Bias Mitigation

7. Pilot Testing

Three-Phase Pilot Testing Protocol:

## Phase 1: Expert Review (n = 3-5 experts)

### Content Validity
- Do items adequately cover the construct?
- Are any important facets missing?
- Are items relevant to the target population?
- Content Validity Index (CVI): Rate each item as
  1 = Not relevant, 2 = Somewhat relevant, 3 = Quite relevant, 4 = Highly relevant
  Item-CVI = proportion of experts rating 3 or 4 (threshold: ≥ 0.78)
  Scale-CVI/Ave = mean of Item-CVIs (threshold: ≥ 0.90)

### Face Validity
- Do items appear to measure what they claim?
- Is the language clear and appropriate?
- Is the survey length reasonable?

---

## Phase 2: Cognitive Interviews (n = 5-10 from target population)

### Think-Aloud Protocol
"Please read each question out loud and tell me what you are thinking
as you decide on your answer."

### Probing Questions
- "What does [term] mean to you?"
- "How did you arrive at your answer?"
- "Was this question easy or difficult to answer? Why?"
- "Can you put this question in your own words?"
- "Is there anything confusing about this question?"
- "Would you change anything about this question?"

### Document
- Items that cause confusion or hesitation
- Items interpreted differently than intended
- Items where response options do not fit
- Suggested wording improvements
- Time to complete each section

---

## Phase 3: Quantitative Pilot (n = 30-50 from target population)

### Assess
- [ ] Completion rate and completion time
- [ ] Item-level missing data (flag items with >10% missing)
- [ ] Response distributions (flag items with >90% in one category)
- [ ] Internal consistency (Cronbach's alpha per subscale)
- [ ] Item-total correlations (flag items < 0.30)
- [ ] Inter-item correlations (flag pairs > 0.85 — redundancy)
- [ ] Open-ended feedback on survey experience
- [ ] Technical issues (display, skip logic, mobile compatibility)

8. Instrument Validation

Reliability

Internal Consistency:

import pandas as pd
import numpy as np

def cronbachs_alpha(df):
    """
    Calculate Cronbach's alpha for a set of items.
    df: DataFrame where each column is an item and each row is a respondent.
    """
    n_items = df.shape[1]
    item_variances = df.var(axis=0, ddof=1)
    total_variance = df.sum(axis=1).var(ddof=1)

    alpha = (n_items / (n_items - 1)) * (1 - item_variances.sum() / total_variance)
    return alpha

# Example
data = pd.DataFrame({
    'item1': [4, 3, 5, 4, 3, 5, 4, 3, 2, 4],
    'item2': [3, 3, 4, 4, 2, 5, 4, 3, 3, 4],
    'item3': [4, 4, 5, 3, 3, 4, 5, 2, 3, 5],
    'item4': [3, 2, 4, 4, 3, 5, 4, 3, 2, 3],
})

alpha = cronbachs_alpha(data)
print(f"Cronbach's alpha: {alpha:.3f}")

# Interpretation:
# α ≥ 0.90  Excellent (but check for redundancy)
# 0.80 ≤ α < 0.90  Good
# 0.70 ≤ α < 0.80  Acceptable
# 0.60 ≤ α < 0.70  Questionable
# α < 0.60  Poor — revise items

Item-Total Correlations:

def item_total_correlations(df):
    """Calculate corrected item-total correlations."""
    results = {}
    for col in df.columns:
        rest = df.drop(columns=col).sum(axis=1)
        corr = df[col].corr(rest)
        results[col] = round(corr, 3)
    return results

itc = item_total_correlations(data)
for item, corr in itc.items():
    flag = " ← REVIEW" if corr < 0.30 else ""
    print(f"  {item}: r = {corr}{flag}")

Validity

Exploratory Factor Analysis (EFA):

from factor_analyzer import FactorAnalyzer
from factor_analyzer.factor_analyzer import calculate_bartlett_sphericity, calculate_kmo

# Check suitability for factor analysis
chi_square, p_value = calculate_bartlett_sphericity(data)
print(f"Bartlett's test: χ² = {chi_square:.2f}, p = {p_value:.4f}")
# p < 0.05 → suitable for factor analysis

kmo_all, kmo_model = calculate_kmo(data)
print(f"KMO: {kmo_model:.3f}")
# KMO > 0.60 → suitable; > 0.80 → good; > 0.90 → excellent

# Determine number of factors (parallel analysis)
fa = FactorAnalyzer(rotation=None, n_factors=data.shape[1])
fa.fit(data)
eigenvalues, _ = fa.get_eigenvalues()
print("Eigenvalues:", [f"{ev:.3f}" for ev in eigenvalues])
# Retain factors with eigenvalue > 1 (Kaiser criterion)
# Also use scree plot and parallel analysis

# Run EFA with chosen number of factors
fa = FactorAnalyzer(n_factors=2, rotation='oblimin', method='ml')
fa.fit(data)

# Factor loadings
loadings = pd.DataFrame(
    fa.loadings_,
    index=data.columns,
    columns=[f'Factor {i+1}' for i in range(2)]
)
print("\nFactor Loadings:")
print(loadings.round(3))
# Items should load ≥ 0.40 on one factor and < 0.30 on others

9. Online Survey Platform Comparison

10. Interview Protocol Development

Semi-Structured Interview Guide Template:

## Interview Protocol

### Study: [Title]
### Interviewer: [Name]
### Participant ID: ___  Date: ___  Start Time: ___

---

### Opening (5 minutes)
- Thank participant for their time
- Review consent (confirm recording permission)
- Explain purpose: "I'm interested in understanding your experiences with [topic]"
- Explain format: "I have some questions prepared, but this is a conversation.
  There are no right or wrong answers. Please share as much or as little
  as you're comfortable with."

### Warm-Up Question (5 minutes)
1. "Can you tell me about your role and how you came to it?"
   - Probe: "How long have you been in this position?"

### Main Questions (30-40 minutes)

**Block A: [Topic/Construct 1]**
2. "Describe your experience with [topic]."
   - Probe: "Can you give me a specific example?"
   - Probe: "How did that make you feel?"
   - Probe: "What happened next?"

3. "What challenges have you encountered related to [topic]?"
   - Probe: "How did you handle that?"
   - Probe: "What support, if any, did you receive?"

**Block B: [Topic/Construct 2]**
4. "How has [topic] changed over time for you?"
   - Probe: "What prompted that change?"
   - Probe: "Looking back, what would you have done differently?"

5. "What factors have been most influential in shaping your [topic]?"
   - Probe: "Can you elaborate on [specific factor mentioned]?"

**Block C: [Topic/Construct 3]**
6. [Question]
   - Probes

### Closing (5 minutes)
7. "Is there anything else about [topic] that you think is important
   and that I haven't asked about?"
8. "Do you have any questions for me?"

### Post-Interview
- Thank participant; explain next steps and timeline
- Stop recording
- Write field notes immediately after:
  - Key impressions
  - Non-verbal observations
  - Reflections on the interview process
  - Emerging analytical ideas

End Time: ___  Total Duration: ___

11. Focus Group Facilitation

Focus Group Design Checklist:

## Focus Group Planning

### Composition
- Participants per group: 6-10 (4-6 for complex topics)
- Number of groups: 3-5 per population segment (until saturation)
- Homogeneity within groups (shared experience/characteristic)
- Heterogeneity across groups (variation in perspectives)

### Roles
- Moderator: Facilitates discussion, manages dynamics
- Note-taker: Records non-verbal cues, group dynamics, key quotes
- Optional: Observer behind one-way glass or via video

### Environment
- Comfortable, neutral, private setting
- Circular or U-shaped seating (no head of table)
- Recording equipment tested before session
- Refreshments available
- Name tents (first names or pseudonyms)

### Facilitation Techniques
- Opening: Icebreaker or round-robin introduction
- Funnel approach: Broad → specific questions
- Manage dominant voices: "Let's hear from others..."
- Draw out quiet participants: "We haven't heard from everyone yet..."
- Handle conflict: "It sounds like there are different perspectives here,
  and that's valuable. Let's explore both views."
- Closing: "Of everything we've discussed, what stands out as most important?"

12. Cultural Adaptation of Instruments

Brislin's (1970) Back-Translation Method:

Original instrument (Source Language)
         │
         ▼
Forward translation by Translator A
(Source → Target language; bilingual with target as dominant)
         │
         ▼
Back-translation by Translator B
(Target → Source language; bilingual with source as dominant;
 has NOT seen original instrument)
         │
         ▼
Compare original and back-translation
(Research team + both translators)
         │
         ├── Discrepancies? → Revise target version → Re-translate → Compare again
         │
         └── Equivalent? → Proceed to expert review
                              │
                              ▼
                    Cultural review panel
                    (Experts familiar with target culture)
                              │
                              ▼
                    Cognitive interviews in target population
                              │
                              ▼
                    Pilot test and psychometric validation
                    in target population

ISPOR Guidelines for Cross-Cultural Adaptation:

1. Preparation (permissions, concept elaboration)
2. Forward translation (2 independent translators)
3. Reconciliation of forward translations
4. Back-translation
5. Back-translation review
6. Harmonization across language versions
7. Cognitive debriefing with target population
8. Review of cognitive debriefing results
9. Proofreading
10. Final report documenting all decisions

Best Practices

1. Start with constructs, not questions. Define exactly what you are measuring before writing a single item. Each item should trace back to a specific construct or dimension.

2. Use existing validated instruments when possible. Do not reinvent the wheel. Search the literature for instruments with established psychometric properties.

3. Pilot everything. Every survey should go through cognitive interviews and a quantitative pilot before full deployment. There is no substitute for testing with your target population.

4. Keep it short. Every additional item increases dropout risk. Include only items you will actually analyze. A good survey is as short as possible and as long as necessary.

5. Design for your weakest respondent. Write at an appropriate reading level. Test on mobile devices. Consider accessibility (screen readers, color contrast). Provide translations if needed.

6. Randomize item order within sections. This reduces order effects and helps detect careless responding.

7. Include attention checks. Embed 1-2 instructed response items (e.g., "Please select 'Agree' for this item") to identify careless respondents.

8. Plan your analysis before collecting data. Every question should have a purpose in your analysis plan. If you cannot say how you will analyze an item, remove it.

9. Document everything. Keep a survey design log recording every decision: why items were added, removed, or revised; pilot test results; expert feedback.

10. Protect respondent data. Use anonymous links when possible; store data securely; minimize collection of identifiers; comply with IRB requirements.

Common Pitfalls

References

• DeVellis, R. F., & Thorpe, C. T. (2022). Scale development: Theory and applications (5th ed.). Sage.
• Dillman, D. A., Smyth, J. D., & Christian, L. M. (2014). Internet, phone, mail, and mixed-mode surveys: The tailored design method (4th ed.). Wiley.
• Fowler, F. J. (2014). Survey research methods (5th ed.). Sage.
• Groves, R. M., Fowler, F. J., Couper, M. P., Lepkowski, J. M., Singer, E., & Tourangeau, R. (2009). Survey methodology (2nd ed.). Wiley.
• Krosnick, J. A., & Presser, S. (2010). Question and questionnaire design. In P. V. Marsden & J. D. Wright (Eds.), Handbook of survey research (2nd ed., pp. 263-313). Emerald.
• Podsakoff, P. M., MacKenzie, S. B., Lee, J. Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research. Journal of Applied Psychology, 88(5), 879-903.
• Willis, G. B. (2005). Cognitive interviewing: A tool for improving questionnaire design. Sage.

See also: references/survey-methodology.md for expanded methodology details.