Reliability in Research: How to Ensure Consistent Results
Imagine stepping on your bathroom scale three times in one minute. The first reading says 70 kg. The second says 85 kg. The third says 62 kg. Would you trust that scale? Of course not. A scale that gives wildly different readings is useless, no matter how expensive or fancy it looks.
The same principle applies to your research instrument. If your questionnaire produces different results every time you administer it—or if different people using it get different answers—your data cannot be trusted. Your supervisor will reject it. Your findings will be meaningless.
This is where reliability comes in.
Reliability is the consistency of your measurement. A reliable research instrument produces the same results under the same conditions, every time. Without reliability, your thesis has no foundation.
In this guide, I will explain what reliability means in research, the four types you need to know, how to measure them using tools like Cronbach’s alpha in SPSS, and practical strategies to ensure your own research produces consistent, trustworthy results.
Struggling with research methodology or data analysis for your thesis? At Proposal Writers Kenya, we help students design reliable instruments, analyze data using SPSS, and write methodology chapters that impress supervisors. Get expert help with your thesis today.
What Is Reliability in Research?
Reliability is the degree to which a research instrument produces consistent and stable results when administered repeatedly under the same conditions.
Think of it this way: If you give the same questionnaire to the same person twice (assuming nothing about them has changed), you should get very similar scores both times. If their scores bounce around randomly, your instrument is unreliable.
Here is a distinction every student must understand:
Reliability is about consistency: Does your instrument give the same result every time?
Validity is about accuracy: Does your instrument measure what it claims to measure?
A measurement can be reliable without being valid. A scale that consistently reads 5 kg too high is reliable (same error every time) but not valid (not accurate). However, a measurement cannot be valid without being reliable. Accuracy depends on consistency.
The Four Types of Reliability
Different research situations call for different types of reliability. Here are the four you need to know for your thesis.
1. Test-Retest Reliability
Test-retest reliability measures whether your instrument produces the same results when administered to the same people at two different points in time.
How it works: You give your questionnaire to a group of participants. Two to four weeks later, you give the exact same questionnaire to the same participants. Then you calculate the correlation between the first and second scores.
When to use it: When you are measuring stable traits that should not change over time, such as personality, intelligence, or attitudes.
Acceptable threshold: A correlation of 0.70 or higher indicates good test-retest reliability.
Limitations: Participants may remember their answers from the first time (practice effect), and true scores can change over time (maturation).
2. Inter-Rater Reliability
Inter-rater reliability measures whether two or more independent observers or raters give similar scores when assessing the same thing.
How it works: Two or more raters independently evaluate the same phenomenon using the same scoring rubric. Then you calculate how much they agree.
When to use it: For observational studies, coding open-ended survey responses, grading essays or presentations, or any research involving human judgment.
How to measure: Cohen’s Kappa for categorical ratings, Intraclass Correlation Coefficient (ICC) for continuous scores.
Acceptable threshold: Kappa above 0.70, ICC above 0.75.
Real example: Two supervisors marking the same thesis defense using a standardized rubric should give similar scores. If one gives 85% and the other gives 45%, your scoring rubric has poor inter-rater reliability.
3. Internal Consistency Reliability (Most Common)
Internal consistency measures whether all the questions on your questionnaire that are supposed to measure the same construct produce similar responses.
How it works: You examine whether participants who agree with Question 1 also tend to agree with Question 2, Question 3, and so on. If your questions are all measuring the same thing, responses should be correlated.
When to use it: For almost every questionnaire-based study in Kenya. If you are using a Likert scale or multiple questions to measure a single concept (like “customer satisfaction” or “job stress”), you need internal consistency.
How to measure: Cronbach’s alpha (the most common method in social science research).
Acceptable thresholds:
α > 0.90: Excellent
α > 0.80: Good
α > 0.70: Acceptable
α < 0.70: Questionable (needs revision)
What affects Cronbach’s alpha: The number of items on your scale (more items = higher alpha) and the strength of correlation between items.
4. Parallel Forms Reliability
Parallel forms reliability measures whether two different versions of the same test produce similar scores.
How it works: You create two equivalent versions of your questionnaire (different questions but measuring the same constructs). You give both versions to the same participants and correlate the scores.
When to use it: When you need to prevent cheating or create pre-test and post-test versions that cannot be identical.
Limitations: Creating two truly equivalent versions is difficult and time-consuming.
How to Measure Reliability in Your Research
Calculating Cronbach’s Alpha in SPSS (Step-by-Step)
Cronbach’s alpha is the most common reliability statistic you will use. Here is exactly how to calculate it in SPSS.
Step 1: Enter your questionnaire data. Each question should be a separate column (variable). Each row is one participant.
Step 2: Click Analyze > Scale > Reliability Analysis.
Step 3: Move all the questions that measure the same construct into the “Items” box. Important: Do not put all your questionnaire questions together unless your entire questionnaire measures a single construct. Calculate alpha separately for each construct.
Step 4: Click “Statistics” and check the following boxes:
“Item” (to see each question’s properties)
“Scale if item deleted” (to see whether removing a question would improve reliability)
Step 5: Click Continue, then OK.
Step 6: Read your output. Look for the “Cronbach’s Alpha” value in the “Reliability Statistics” table.
Understanding “Cronbach’s Alpha if Item Deleted”
This table tells you whether any question is hurting your reliability. Look at the column “Cronbach’s Alpha if Item Deleted.”
If removing a question would increase your alpha, consider deleting that question
If removing a question would decrease your alpha, keep it
Important: Do not delete questions just to increase alpha. If a question is theoretically important to your construct, keep it even if it slightly lowers alpha. Report both the alpha value and your decision.
Practical Strategies to Improve Reliability
Before Data Collection
| Strategy | How to Implement |
|---|---|
| Write clear, unambiguous questions | Avoid double-barreled questions, jargon, and vague terms. Each question should ask about one thing only. |
| Use established instruments | Adapt validated questionnaires from previous studies rather than creating everything from scratch. |
| Train your research assistants | Ensure everyone administering your instrument follows the exact same procedures. |
| Standardize data collection | Create a written protocol for every step of data collection. |
| Conduct pilot testing | Identify problematic questions before real data collection. |
| Use multiple items per construct | Never measure a complex construct with a single question. Use at least 3-4 questions per construct. |
During Data Collection
Maintain consistent conditions: Administer questionnaires in similar environments (quiet rooms, similar times of day)
Train observers thoroughly: For observational studies, practice until agreement between observers is high
Document everything: Record any deviations from your protocol
After Data Collection
Calculate and report reliability coefficients for your actual data
Average multiple observations (for inter-rater reliability, using averaged scores improves reliability)
Acknowledge limitations in your thesis if reliability is lower than expected
Common Mistakes Kenyan Students Make
| Mistake | Why It Is Wrong | How to Fix |
|---|---|---|
| Ignoring reliability entirely | Your supervisor will question your results | Always calculate and report reliability |
| Reporting Cronbach’s alpha for the entire questionnaire | Alpha assumes all questions measure one construct | Calculate alpha separately for each construct |
| Thinking reliability equals validity | They measure different things | Understand and address both |
| Deleting items just to increase alpha | Removing theoretically important items harms validity | Keep theoretically important items even if alpha drops slightly |
| Collecting pilot data but not calculating reliability | You miss the most important diagnostic | Always run reliability analysis on pilot data |
How to Report Reliability in Your Thesis
Your methodology chapter must include a reliability section. Here is a sample paragraph you can adapt for your own thesis:
“Internal consistency reliability was assessed using Cronbach’s alpha coefficient. The questionnaire measured four constructs: perceived usefulness (6 items), ease of use (5 items), attitude (4 items), and behavioral intention (3 items). Reliability analysis revealed acceptable internal consistency for all constructs: perceived usefulness (α = 0.87), ease of use (α = 0.82), attitude (α = 0.79), and behavioral intention (α = 0.84). These values exceed the recommended threshold of 0.70 (Nunnally, 1978), indicating that the questionnaire produces consistent responses.”
For observational studies:
“Inter-rater reliability was assessed using Cohen’s Kappa. Two independent observers rated 20 randomly selected sessions using the observation checklist. The Kappa coefficient was 0.81, indicating excellent agreement between raters according to Landis and Koch’s (1977) criteria.”
Frequently Asked Questions
What is an acceptable Cronbach’s alpha for an undergraduate thesis?
Alpha above 0.70 is acceptable. For undergraduate work, some supervisors accept 0.65. Below 0.60 is problematic.
Can I calculate reliability if I only have one question per construct?
No. Cronbach’s alpha requires at least two items. If you have single-item measures, you cannot calculate internal consistency. Consider test-retest reliability instead.
My Cronbach’s alpha is 0.60. What should I do?
Review the “Cronbach’s Alpha if Item Deleted” table. Remove the worst-performing question and recalculate. If still low, you may need to add more questions or revise your instrument.
Do I need to report reliability for qualitative research?
The concept of reliability applies differently to qualitative research. Instead of statistical coefficients, qualitative researchers discuss trustworthiness, dependability, and confirmability. Ask your supervisor for guidance.
What is the difference between Cronbach’s alpha and KR-20?
Cronbach’s alpha is for continuous or Likert-scale data. KR-20 (Kuder-Richardson Formula 20) is for binary data (correct/incorrect, yes/no). Most social science research uses Cronbach’s alpha.
Conclusion
Reliability is the foundation of trustworthy research. Without it, your findings cannot be replicated, your conclusions cannot be defended, and your thesis will struggle to gain approval.
The good news is that reliability is entirely within your control. Write clear questions. Use multiple items for each construct. Pilot test your instrument. Calculate Cronbach’s alpha. Report your results transparently.
These steps take time, but they separate a passing thesis from an excellent one. Supervisors notice when students understand and address reliability. Your methodology chapter will stand out.
Remember: reliability is necessary but not sufficient. You also need validity. But you cannot have validity without reliability first. Start with consistency. Build from there.
Need expert help with research methodology, instrument design, or data analysis? At Proposal Writers Kenya, our team of methodology specialists can help you design reliable instruments, calculate Cronbach’s alpha in SPSS, and write a methodology chapter that impresses your supervisor. From undergraduate to PhD level, we support Kenyan students at every stage of their research journey. Visit Proposal Writers Kenya to learn how we can help you succeed.
Ready to get started? Get a free quote for expert thesis support today.