IB Biology IA Statistics: t-Tests, Error Bars and R²

IB Biology IA statistics explained

Effective IB Biology IA statistics should help answer your research question, not merely make the report appear sophisticated. You need to summarize variation, display the data clearly, choose a statistical test that matches the design, check its assumptions, and interpret the result biologically.

Under the current course, first assessed in 2025, the Biology IA is officially called the scientific investigation. It contributes 20% of the final grade, has a maximum of 3,000 words, and requires students to gather and analyse data in response to their own research question. The IB does not prescribe one statistical test for every investigation, so relevance and correct interpretation matter more than complexity.

What statistics need to accomplish in a Biology IA

The current scientific investigation is assessed using four criteria: Research design, Data analysis, Conclusion, and Evaluation. Statistical work primarily supports Data analysis, but its interpretation also strengthens the Conclusion and Evaluation criteria.

A coherent analysis usually does the following:

1. Presents relevant raw data with units and measurement uncertainties.
2. Processes the data using suitable quantities such as means, rates, percentages, or standard deviations.
3. Displays patterns using an appropriate graph.
4. Applies a justified statistical test when it can answer the research question.
5. Interprets the result alongside biological reasoning, uncertainty, and limitations.

There is no official rule that every Biology IA must contain a t-test, error bars, or an R² value. A test is useful only when it matches the variables, sampling method, and hypothesis. Adding an inappropriate test can weaken the analysis because it suggests that the student does not understand the design.

Start by identifying the IA design

Before selecting a test, classify your variables and determine whether observations are independent.

An independent observation comes from a separate experimental unit. Measuring five leaves from one plant does not necessarily provide five independent plant replicates. Treating subsamples as independent replicates is pseudoreplication, which artificially increases sample size and can produce misleading p-values.

Using a t-test in the Biology IA

A t-test evaluates whether a difference between two means is large relative to the variation in the data. Its null hypothesis normally states that the relevant population means, or the mean paired difference, are equal.

Independent or paired t-test?

Use an independent-samples t-test when the measurements in one group do not correspond to measurements in the other. For example, seedlings randomly assigned to two salt concentrations form independent groups.

Use a paired t-test when each value has a meaningful partner. This includes before-and-after measurements on the same organism or matched experimental units. The paired test analyses the differences within pairs, not the two sets as unrelated groups.

For independent groups, Welch's t-test is often a sensible default because it does not assume equal population variances. A conventional pooled t-test requires the additional assumption of equal variances.

Assumptions to consider

A t-test generally requires:

• a continuous dependent variable
• independent experimental units
• no extreme, unjustified outliers
• approximately normal data within groups, or approximately normal paired differences for a paired test
• equal variances only when using the pooled independent t-test

Small samples make normality difficult to evaluate, so inspect dot plots, histograms, or residuals rather than relying mechanically on a normality test. If the distribution is strongly skewed or ordinal, a non-parametric alternative such as Mann-Whitney U or Wilcoxon signed-rank may be more defensible.

Worked t-test interpretation

Suppose an investigation compares mean amylase reaction rate at two temperatures:

A two-tailed Welch's t-test produces t = 3.45, df = 13.8, p = 0.004. A suitable statement is:

The mean reaction rate was higher at 30 °C than at 20 °C, with a difference of 1.3 arbitrary units min⁻¹. Welch's t-test indicated evidence against the null hypothesis, t(13.8) = 3.45, p = 0.004. Within the investigated range, this supports the prediction that increased temperature raised amylase activity, although the result does not establish the complete temperature-response curve.

Do not write that there is a “96% chance the alternative hypothesis is true.” A p-value is the probability of obtaining a result at least this extreme, assuming the null hypothesis and statistical model are valid. Report the exact p-value where software provides it, alongside group means, SDs, sample sizes, and the direction of the difference.

Standard deviation, uncertainty, and error bars

These terms describe different aspects of data quality and should not be used interchangeably.

The sample standard deviation is

s = √[Σ(xᵢ − x̄)²/(n − 1)].

A large SD means the observations are widely dispersed. It does not automatically mean that the measuring instrument was inaccurate. Variation may result from genuine biological differences, inconsistent control variables, measurement limitations, or a mixture of these causes.

Choosing and interpreting error bars

Error bars must be identified in the graph caption or legend. For example, write “points show mean ± 1 SD, n = 6 independent samples,” rather than simply “error bars included.”

Use SD bars when you want to show the spread among observations. Use 95% confidence intervals when you want to communicate the precision of an estimated mean. SE bars are permissible when correctly labelled, but they are smaller than SD bars and should not be presented as if they show the spread of the raw data.

Do not decide statistical significance by looking only at whether error bars overlap. The meaning of overlap depends on whether the bars show SD, SE, or confidence intervals, as well as sample size and study design. Use the appropriate hypothesis test for a formal comparison.

Measurement uncertainty and SD may both appear in an IA because they answer different questions. A balance uncertainty of ±0.01 g describes instrumental resolution, while the SD of repeated plant masses describes variation among the plants or trials.

Correlation, regression, and R²

Use correlation when the research question asks whether two measured variables vary together. A scatter plot should come first because it reveals non-linearity, clusters, and outliers that a single coefficient can conceal.

Pearson's r and Spearman's rank

Pearson's correlation coefficient, r, measures the direction and strength of a linear relationship between two continuous variables. It is most suitable when the relationship is approximately linear, observations are independent, serious outliers are absent, and the assumptions needed for inference are reasonable.

Spearman's rank correlation, rₛ, evaluates a monotonic relationship using ranks. It is useful for ordinal data or continuous data that do not satisfy Pearson's assumptions. Neither coefficient proves causation because confounding variables or reverse causation may explain an association.

What R² means

For an ordinary regression, R², the coefficient of determination, is the proportion of variation in the dependent variable accounted for by the fitted model. If a linear model gives R² = 0.81, then 81% of the observed variation in the dependent variable is accounted for by that fitted linear relationship in the sample.

R² does not show the direction of a relationship. A positive and a negative linear relationship can have the same R². It also does not prove causation, validate the experimental method, or show that the selected model is biologically appropriate.

For simple linear regression with an intercept, R² equals Pearson's r². This equivalence should not be transferred uncritically to Spearman's rank coefficient or to every regression model. Examine the scatter plot and residual pattern rather than treating a high R² as sufficient evidence of a good model.

A practical report might state: “Linear regression gave R² = 0.78, indicating that the model accounted for 78% of the variation in oxygen production rate. However, the residuals showed curvature at high light intensities, suggesting that a linear model did not represent saturation across the entire range.”

When to use chi-square

A chi-square test is for frequencies or counts, not means, percentages derived from tiny samples, or continuous measurements. A goodness-of-fit test compares observed counts with a specified expected distribution, while a test of independence examines association between two categorical variables.

Suppose an F₂ cross produces 32 purple and 18 white flowers, with an expected 1:1 ratio. The expected counts are 25 and 25, so:

χ² = (32 − 25)²/25 + (18 − 25)²/25 = 3.92.

With df = 1, the p-value is approximately 0.048. At α = 0.05, this provides evidence against the 1:1 null model, but the conclusion should also discuss sampling, classification errors, and whether the expected ratio was biologically justified. Expected counts should generally not be very small; combining categories or using an exact method may be necessary when chi-square assumptions are not met.

A practical workflow for statistical analysis

1. State the biological hypothesis and null hypothesis. Define the population or process being investigated.
2. Identify the experimental unit. Separate independent replicates from repeated readings or subsamples.
3. Show the raw data. Include units, uncertainty, consistent precision, and relevant qualitative observations.
4. Calculate descriptive statistics. Give sample size, mean or median, and SD or another justified measure of spread.
5. Graph individual data where possible. Scatter plots, dot plots, or box plots often reveal more than bars alone.
6. Select and justify the test. Link the choice to variable type, number of groups, pairing, and assumptions.
7. Report the output transparently. Include the test statistic, degrees of freedom where relevant, p-value, direction, and descriptive results.
8. Interpret rather than restate. Connect the statistical result to the research question, biological mechanism, uncertainty, and study limitations.

You can compare your structure with RevisionDojo Biology IA examples, the worked mung bean IA exemplar, and the guide to statistical analysis in a science IA. These should be used to study analytical decisions, not as templates to copy.

Common statistical mistakes

• Selecting a test after seeing which one gives p < 0.05.
• Running several t-tests across three or more groups instead of considering ANOVA.
• Treating repeated readings from one organism as independent biological replicates.
• Reporting R² without showing the scatter plot or checking model fit.
• Calling SD “experimental uncertainty” without explaining what varied.
• Leaving error bars unidentified.
• Claiming that p > 0.05 proves there is no effect.
• Removing an outlier only because it weakens the preferred conclusion.
• Reporting calculator output without units, context, or biological interpretation.

An outlier should be retained unless there is a transparent methodological or statistical justification for exclusion. If you exclude one, state the rule, report its effect on the analysis, and explain the decision rather than silently deleting the value.

Conclusion

Strong IB Biology IA statistics begin with the design. Use a t-test for an appropriate two-mean comparison, chi-square for categorical counts, and correlation or regression for relationships between variables. Distinguish measurement uncertainty from SD, label every error bar, and interpret R² as model fit rather than proof of causation.

RevisionDojo's Biology IA grader can help you review whether your analysis is clearly presented and interpreted. For exam preparation after completing the IA, use the IB Biology past paper video solutions to practise data-based questions and compare your reasoning with worked explanations.

Sources and referenced URLs

• IB Sciences: Biology subject brief, first assessment 2025
• International Baccalaureate Biology curriculum updates
• NIST paired-sample t-test guidance
• NIST critical values of the chi-square distribution
• GraphPad guidance on choosing SD, SEM, or confidence intervals
• GraphPad guidance on error-bar overlap and significance
• Error bars in experimental biology
• RevisionDojo statistical analysis for an IB Science IA
• RevisionDojo Biology IA examples
• RevisionDojo mung bean Biology IA exemplar
• RevisionDojo Biology IA grader
• RevisionDojo IB Biology past paper video solutions