Quantitative Research Methods Guide | Survey Design, Data Analysis, and SPSS Application

What this page helps you do first

• Core elements of questionnaire design: scale selection, questionnaire items, pilot testing, reliability and validity
• Sample size calculation methods and common statistical formulas
• Complete analysis process: descriptive statistics → reliability → hypothesis testing

What is quantitative research and its core differences from qualitative research

Quantitative research collects quantifiable data and uses statistical methods to test research hypotheses. Its core characteristics: representing concepts with numbers, verifying relationships through statistical analysis, pursuing generalizability of conclusions.

Compared to qualitative research, quantitative research is more suitable for answering questions like "how much," "what relationships," "how large is the impact." For example: what is the user satisfaction level? How strong is the influence of price on purchase intention? Are behavioral differences across groups statistically significant?

Core elements and steps of questionnaire design

• Step 1—Operationalization of research variables: transform abstract concepts (like "user satisfaction") into measurable specific indicators (like "product satisfaction," "service satisfaction," "value-for-money satisfaction")
• Step 2—Scale selection: prioritize established scales (like Likert scales), which have literature backing and guaranteed reliability and validity
• Step 3—Questionnaire item design: at least 3 items per variable, avoid double negatives and multi-meaning items
• Step 4—Pilot testing: conduct pre-testing with 20-50 people before formal distribution to test questionnaire comprehensibility and response time
• Step 5—Formal distribution and collection: ensure sample size meets statistical testing requirements, response rate generally should exceed 70%

Sample size calculation methods and statistical requirements

• Rule of thumb: sample size at least 10-20 times the number of variables. If study contains 20 variables, sample size should reach 200-400
• Cochran formula: n = Z²×p×(1-p)/e², where Z is confidence level coefficient (1.96 for 95% confidence), p is expected proportion, e is margin of error
• Stratified sampling sample size: for stratified sampling, internal sample size of each layer needs to meet analysis requirements
• Note: invalid questionnaires (incomplete, pattern responses) should not exceed 20% of collected questionnaires; otherwise increase distribution quantity
• Sample size is not "the bigger the better"—under correct statistical analysis, the minimum sample size meeting requirements is sufficient

Complete SPSS analysis process

• Step 1—Data entry and cleaning: check missing values, outliers, patterned responses, ensure data quality
• Step 2—Descriptive statistics: mean, standard deviation, frequency distribution, normality testing (K-S test)
• Step 3—Reliability analysis: Cronbach's α coefficient >0.7 indicates acceptable internal consistency, >0.8 indicates good reliability
• Step 4—Validity analysis: KMO value >0.7 suitable for factor analysis, Bartlett's sphericity test significant (p<0.05) indicates correlation between variables
• Step 5—Correlation analysis: Pearson correlation coefficient determines direction and strength of variable relationships
• Step 6—Regression analysis: choose linear regression (continuous variables) or Logistic regression (binary variables) based on dependent variable type

Regression analysis types and application scenarios

• Linear regression: dependent variable is continuous (satisfaction scores, purchase intention scores). Output includes R² (explanatory power), F-test (model significance), β coefficient (impact degree)
• Logistic regression: dependent variable is binary (purchase or not, churn or not). Output is OR value (odds ratio) and significance level
• Hierarchical regression: testing moderating or mediating effects, stepwise adding control variables, independent variables, moderating variables in regression
• Note: regression requires no severe multicollinearity among independent variables (VIF<10)—conduct multicollinearity diagnosis before analysis

Common mistakes in quantitative research papers and analysis

• Mistake 1: Inappropriate scale selection → Using unvalidated self-developed scales lacking reliability and validity evidence. Improvement: prioritize established scales and report their sources
• Mistake 2: Insufficient sample size → Small sample leads to insufficient statistical power, unable to detect truly existing effects. Improvement: conduct sample size planning in advance
• Mistake 3: Misinterpreting significance → Misunderstanding statistical significance as practical significance. Improvement: focus on effect size rather than just p-value
• Mistake 4: Over-causal inference → Interpreting correlations as causal relationships. Improvement: clearly explain limitations in discussion, avoid over-inference

Frequently asked questions

Must questionnaires use Likert 5-point scales?

Not necessarily. Likert scales have variations like 5-point, 7-point, 10-point. More points enhance respondent discrimination but increase difficulty. 5-point and 7-point are most common in academic research. Consider: respondent characteristics (general consumers better suit 5-point), research precision requirements, and comparability with existing research.

How many samples for pilot testing?

Pilot testing sample size is typically 20-50 people, aiming to test questionnaire comprehensibility and identify obvious problems. Pilot testing does not need to meet formal analysis statistical requirements—focus is on discovering questionnaire design flaws like item ambiguity and logical skip issues.

What if reliability and validity tests do not pass?

Reliability below standard (α<0.7): check if reverse-scored items are properly reversed, if items match variable concepts, consider deleting or modifying non-compliant items. Validity below standard (KMO<0.7): check if sample size is sufficient and item correlations are adequate; increase sample or adjust items if necessary.

How to handle high VIF values in regression analysis?

VIF>10 indicates severe multicollinearity. Solutions: 1) remove redundant variables (keep one of two highly correlated independent variables); 2) use principal component regression or partial least squares regression; 3) increasing sample size may alleviate but not fundamentally solve multicollinearity.

Does non-significant p-value mean the hypothesis is wrong?

Not necessarily. Non-significant p-value can occur for two reasons: insufficient sample size leading to low statistical power, or truly no relationship existing. Should judge combined with effect size: if effect size is very small, even if significant it has little practical meaning; if effect size is large but p-value is non-significant, consider increasing sample size.