শিক্ষামূলক নোট: এই পৃষ্ঠা একাডেমিক জীববিজ্ঞান শেখা ও পরীক্ষার প্রস্তুতির সহায়ক।
T-test: Significant Difference Between Means
Concept Overview
Student’s t-test হলো এমন একটি inferential statistical test, যা দুইটি mean-এর পার্থক্য random sampling error দিয়ে ব্যাখ্যা করা যায় কি না তা যাচাই করে। Biology, medicine, agriculture and ecology-তে অনেক সময় sample size ছোট হয় এবং population standard deviation অজানা থাকে। এই অবস্থায় t-test গবেষককে evidence-based decision নিতে সাহায্য করে।
Core idea:
Observed mean difference
↓
Compare with standard error
↓
Calculate t-value
↓
Use degrees of freedom
↓
Interpret p-value
↓
Accept or reject null hypothesis with caution
Why This Matters
দুইটি গ্রুপের mean আলাদা দেখালেই তা biological effect নয়। যেমন treated plant-এর average height control plant-এর চেয়ে বেশি হতে পারে, কিন্তু sample variation বেশি হলে সেই difference statistically reliable নাও হতে পারে। t-test শেখায়: numerical difference, sampling variation and evidence strength—এই তিনটি একসাথে বিচার করতে হয়।
T-test Learning Focus
এই lecture central LBFL framework-কে hypothesis testing-এ প্রয়োগ করে। Learner-এর focus হবে null hypothesis, alternative hypothesis, t-value, standard error, degrees of freedom, p-value, assumptions, test selection, and biological interpretation.
Historical Background
William Sealy Gosset 1908 সালে “Student” ছদ্মনামে t-test প্রকাশ করেন। তাঁর কাজের মূল উদ্দেশ্য ছিল ছোট sample নিয়ে নির্ভরযোগ্য সিদ্ধান্ত নেওয়া। Biological research-এ আমরা প্রায়ই ছোট sample পাই—যেমন 10 fish, 12 plants, 15 experimental plots, or paired before-after measurements.
Hypothesis Framework
Null hypothesis — H₀
দুইটি mean-এর মধ্যে কোনো বাস্তব পার্থক্য নেই; observed difference sampling error হতে পারে।
Alternative hypothesis — H₁
দুইটি mean-এর মধ্যে বাস্তব বা statistically meaningful পার্থক্য আছে।
Example question: নতুন feed মাছের ওজন বৃদ্ধি করে কি না?
H₀: নতুন feed এবং পুরনো feed-এর mean weight gain-এ পার্থক্য নেই।
H₁: নতুন feed এবং পুরনো feed-এর mean weight gain-এ পার্থক্য আছে।
Main Types of t-test
| t-test type | When used | Biological example |
|---|---|---|
| One-sample t-test | one sample mean vs known value | species mean length vs published value |
| Independent two-sample t-test | two independent groups | control vs treated plant height |
| Paired t-test | same subject/plot before-after | before-after blood pressure or body mass |
| Welch’s t-test | two independent groups with unequal variance | two habitats with unequal variation |
Key Assumptions
Measurement scale
Data should usually be continuous or measurement-scale.
Approximate normality
Small sample data should not show severe non-normality.
Independent observations
One observation should not improperly influence another.
Variance condition
Equal variance is needed for pooled independent t-test; otherwise Welch's t-test is safer.
Formula: Independent Two-Sample t-Test
t = (X̄₁ − X̄₂) / SEdifference
Where:
SEdifference = √(s₁²/n₁ + s₂²/n₂)
- X̄₁ and X̄₂ = sample means
- s₁² and s₂² = sample variances
- n₁ and n₂ = sample sizes
- SE = standard error of difference
Worked Biological Example
Suppose two rice varieties are compared for yield.
| Group | n | Mean yield | SD |
|---|---|---|---|
| Variety A | 10 | 42 kg | 4 kg |
| Variety B | 10 | 37 kg | 5 kg |
Question: Is the mean yield difference likely to be statistically meaningful?
Calculation logic:
Mean difference = 42 − 37 = 5 kg
SEdifference = √(4²/10 + 5²/10)
= √(16/10 + 25/10)
= √4.1
≈ 2.02
t ≈ 5 / 2.02
≈ 2.47
Interpretation requires degrees of freedom and p-value or critical t-value. The statistical result should then be interpreted with biological context: sample size, measurement quality, experimental design and practical importance.
p-value Interpretation
| Misinterpretation | Better interpretation |
|---|---|
| p-value is the probability that H₀ is true | p-value is probability of observing this result or more extreme if H₀ were true |
| p < 0.05 proves biological importance | p < 0.05 suggests statistical evidence, not automatic biological importance |
| non-significant means no effect exists | sample may be underpowered or variation may be high |
| significant result means study is perfect | design, bias, measurement and effect size still matter |
Statistical vs Biological Significance
Statistical significance
↓
Is the observed difference unlikely under H₀?
Biological significance
↓
Is the difference meaningful in real biological, ecological or health context?
A small difference can be statistically significant in a huge sample but biologically trivial. A large difference can be biologically important but statistically non-significant if sample size is too small or variation is too high.
Test Selection Flow
One group vs known value?
↓ yes
One-sample t-test
Two independent groups?
↓ yes
Independent t-test or Welch's t-test
Same subject before-after?
↓ yes
Paired t-test
Common Mistakes to Avoid
Mistake 1
Using t-test for categorical data. Categorical data often needs chi-square or similar methods.
Mistake 2
Ignoring paired design and using independent t-test for before-after data.
Mistake 3
Reporting p-value without effect size, sample size or biological context.
Mistake 4
Confusing statistical significance with proof of causation.
Synaptic Bridge
T-test teaches disciplined doubt. A visible difference may be real, or it may be noise. In science and life, a careful thinker does not jump from observation to conclusion; they compare evidence against uncertainty. Biostatistics therefore strengthens critical thinking.
Critical Thinking Questions
- Why is a mean difference alone not enough for scientific conclusion?
- When should paired t-test be used instead of independent t-test?
- Why does standard error affect t-value?
- Why can p < 0.05 still be biologically unimportant?
- How can poor experimental design weaken a statistically significant result?
Related Learning Paths
References
- Standard HSC Zoology Biostatistics notes.
- Integrated Zoology and Research Methodology references on hypothesis testing.
- General biostatistics references on Student’s t-test, p-value and biological interpretation.