IB Group 4 Science IA: Statistical Analysis

Mastering statistical analysis is not just a box to tick for your IB Science Internal Assessment (IA); it's the core of your scientific argument. Your IA tells a story with data, and statistics provide the objective language to validate that story, transforming observations into evidence. A well-chosen and correctly interpreted statistical test elevates your work from a simple report to a genuine scientific investigation, demonstrating your ability to handle data with rigor and sophistication.

Understanding Your Data: The Foundation of Test Selection

Before you can choose a statistical test, you must intimately understand your data. Every test is built on specific assumptions about the data it analyzes, and using the wrong test invalidates your entire analysis. You must first categorize your variables—the factors you measure or manipulate.

The most fundamental distinction is between categorical data and continuous data. Categorical data represents groups or labels, such as plant species (oak, maple, pine) or reaction outcome (successful, unsuccessful). Continuous data can take any numerical value within a range, like height (172.5 cm), temperature (22.3°C), or time (45.7 seconds). Furthermore, you must identify which variable is your independent variable (the one you change or categorize) and which is your dependent variable (the one you measure).

Finally, assess the distribution of your continuous data. Many common tests, like the t-test, assume your data is normally distributed—following the classic bell curve. You can check this using tools like histograms, Q-Q plots, or normality tests (e.g., Shapiro-Wilk). If your data is not normal, you may need to use non-parametric tests, which do not assume a specific distribution.

Formulating Hypotheses: The Question Behind the Test

Statistics are used to test a clear, predictive statement. This begins with formulating a null hypothesis (H₀) and an alternative hypothesis (H₁). The null hypothesis is the default position of "no effect" or "no difference." For example, "There is no difference in the mean growth rate of plants under blue light versus red light." The alternative hypothesis is what your experiment is trying to provide evidence for, such as "There is a difference in the mean growth rate..."

Your statistical test will calculate a probability (p-value) based on your data. The p-value represents the probability of obtaining your observed results (or more extreme results) if the null hypothesis were true. A low p-value (typically ≤ 0.05) suggests your data is unlikely under H₀, providing grounds to reject the null hypothesis in favor of H₁. Crucially, you never "accept" a hypothesis; you either reject H₀ or fail to reject it based on the evidence.

The Test Selection Framework: Matching Test to Question

Choosing the correct test is a logical process based on your data type and the question you are asking. Follow this decision framework:

1. Comparing Means: Are you comparing the average (mean) of one group to a standard, or the means of two different groups?

• One-sample t-test: Compares the mean of a single sample to a known standard or theoretical value.
• Unpaired (independent) t-test: Compares the means of two independent, separate groups (e.g., control vs. treatment).
• Paired t-test: Compares means from the same subjects under two different conditions (e.g., heart rate before and after exercise).

1. Testing for Relationships: Are you investigating a connection between two variables?

• Pearson's correlation coefficient (r): Measures the strength and direction of a linear relationship between two continuous variables.
• Spearman's rank correlation coefficient (ρ): Measures a monotonic relationship (whether linear or not) and is used for ordinal data or when data is not normally distributed.

1. Testing for Independence or Goodness-of-Fit: Are you comparing observed categorical frequencies to expected frequencies?

• Chi-squared (χ²) test of independence: Determines if there is a significant association between two categorical variables (e.g., is gender independent of preferred learning style?).
• Chi-squared (χ²) goodness-of-fit test: Determines if the observed distribution of a single categorical variable matches an expected distribution (e.g., do the observed ratios of plant phenotypes match a 3:1 Mendelian ratio?).

Executing and Interpreting Key Tests

The T-Test in Action

Suppose your IA investigates the effect of a fertilizer on plant height. You have a control group (water) and a treatment group (fertilizer solution). You measure the final height of 15 plants in each group. This calls for an unpaired t-test.

1. State Hypotheses: H₀: μcontrol = μtreatment; H₁: μcontrol ≠ μtreatment.
2. Check Assumptions: Data is continuous (height), groups are independent, and data is approximately normally distributed in each group.
3. Calculate: Software will output a t-statistic and a p-value. The t-statistic represents the size of the difference relative to the variation in your data. Let's say you obtain t = 2.85, p = 0.008.
4. Interpret: Since p = 0.008 < 0.05, you reject the null hypothesis. You conclude: "There is a statistically significant difference in mean plant height between the control and treatment groups (t = 2.85, p = 0.008), supporting the hypothesis that the fertilizer affects growth."

The Chi-Squared Test in Action

Imagine an ecology IA where you survey leaf litter invertebrate distribution in sun vs. shade. You count frequencies of three insect orders (Beetles, Spiders, Ants) in each location.

1. State Hypotheses: H₀: Insect order distribution is independent of location (sun/shade). H₁: Distribution is dependent on location.
2. Calculate: Software creates a contingency table and computes the χ² value by summing $(Observed-Expected{)}^2/Expected$ for all cells. It gives you χ² = 10.2 with a corresponding p-value = 0.006.
3. Interpret: p = 0.006 < 0.05, so you reject the null hypothesis. Conclude: "There is a statistically significant association between insect order and microhabitat (χ² = 10.2, p = 0.006), suggesting habitat preference differs among taxa."

Correlation Analysis in Action

For a physics IA on pendulum motion, you might explore the relationship between string length and period.

1. Choose Test: Both variables (length in cm, period in s) are continuous. After checking, the relationship appears linear and data is normal, so use Pearson's r.
2. Calculate: Software outputs r = 0.98, p < 0.001.
3. Interpret: The correlation coefficient r = 0.98 indicates a very strong positive linear relationship. The p-value confirms this correlation is statistically significant. Crucially, you must remember: correlation does not imply causation. You have demonstrated a mathematical relationship, not proven that length causes the change in period (though in this known law, it does).

Presenting Your Statistical Analysis with Clarity

Presentation is key to communicating understanding. Integrate statistics seamlessly:

• In-text: "An unpaired t-test revealed the enzyme reaction rate at 40°C was significantly higher than at 25°C (t = 3.41, p = 0.002)."
• In tables/graphs: Include asterisks ( for p<0.05, * for p<0.01) on graphs to denote significance. Report exact p-values in tables where possible.
• Justification: In your "Data Analysis" section, explicitly justify why you chose a particular test based on your data types and research question. This demonstrates informed decision-making.

Common Pitfalls

1. Using the Wrong Test: The most critical error. Applying a t-test to categorical data or a correlation to compare means renders results meaningless. Correction: Always use the selection framework. Describe your variables first, then choose the test that matches.
2. Misinterpreting the P-Value: A p-value is not the probability that the null hypothesis is true, nor is it the probability your results are due to chance alone. It is the probability of the data given that H₀ is true. Correction: Phrase conclusions carefully: "The data provides sufficient evidence to reject the null hypothesis," not "This proves my hypothesis."
3. Ignoring Test Assumptions: Running a parametric test (like a t-test) on clearly non-normal data or data with unequal variances between groups can give misleading p-values. Correction: Always perform and report checks for normality and homogeneity of variance. If assumptions are violated, use the non-parametric alternative (e.g., Mann-Whitney U test instead of unpaired t-test).
4. Overlooking Practical vs. Statistical Significance: A result can be statistically significant (tiny p-value) but practically meaningless if the effect size is trivial. Correction: Report and comment on the effect size. For a t-test, this could be the difference between the means. For correlation, it's the r-value itself. A large p-value does not mean "no effect," only that you didn't detect one with your sample.

Summary

• Your choice of statistical test is dictated by the type of data you have (categorical vs. continuous) and the specific question you are asking (comparing means, testing relationships, etc.).
• Always begin with clear null and alternative hypotheses. The p-value helps you decide whether to reject H₀, typically using a threshold of 0.05.
• Justify your test selection explicitly in your IA and ensure you have checked the test's assumptions (like normality for t-tests) before proceeding.
• Interpret results in context. State what the statistical finding (e.g., p < 0.05) means for your original research question, using precise language about evidence and significance.
• Present statistical results clearly in-text and on graphs, and always remember that statistical significance should be considered alongside the real-world effect size.