Chapter 12: Data-Based and Statistical Reasoning

Before you can analyze data, you need to read it. The MCAT loves to bury key information inside graphs and tables, then ask you to draw conclusions under time pressure. Think of each figure like a map - if you know where to look and what the symbols mean, you can navigate quickly. If you do not, you will waste precious minutes squinting at axes and legends.

This section covers every data format you will encounter on test day, along with the visual traps the MCAT uses to mislead careless readers.

Tables

Tables are the simplest way to present data - rows and columns of numbers. On the MCAT, tables typically present raw experimental results, patient characteristics, or comparison data between groups.

When reading a table, scan for the largest and smallest values, any obvious patterns (does a value consistently increase down the rows?), and any unexpected outliers. The MCAT rarely asks you to perform complex calculations from a table - it tests whether you can identify trends and make comparisons.

Bar Graphs

Bar graphs compare discrete categories. Each bar represents a different group, and the height shows the measured value. They are ideal for comparing things like average blood pressure across treatment groups or survey responses across age brackets.

Key details to check: Are error bars included? Do the bars start at zero? Are the groups meaningfully different, or do the error bars overlap?

Histograms

Histograms look like bar graphs but represent something fundamentally different. Instead of comparing categories, histograms show the distribution of a single continuous variable. Each bar covers a range of values (a “bin”), and the height shows how many data points fall in that range.

For example, a histogram of exam scores might have bins of 60-70, 70-80, 80-90, and 90-100. The tallest bar tells you where most scores cluster.

Line Graphs

Line graphs display how a variable changes over a continuous range, usually time. Points are connected by lines to show trends. They are the go-to format for showing how a measurement changes during an experiment (drug concentration over hours, population growth over years, reaction rate over temperature).

When interpreting line graphs, focus on: the overall trend (increasing, decreasing, or flat), the slope (steep = rapid change, flat = slow change), and any inflection points where the trend reverses.

Scatter Plots

Scatter plots show the relationship between two continuous variables. Each point represents one observation, plotted by its x-value and y-value. They are used to visualize correlations.

Look for: an upward trend (positive correlation), a downward trend (negative correlation), or a random cloud (no correlation). If a best-fit line is drawn, note its slope and how tightly the points cluster around it.

Pie Charts

Pie charts show proportions of a whole. Each slice represents the fraction of the total belonging to a category. They are less common on the MCAT than other graph types but may appear in passages about demographics, budget allocations, or causes of disease.

The key limitation: pie charts make it hard to compare similar-sized slices. If two slices look nearly the same, you need the actual percentages to tell them apart.

Common MCAT Graph Traps

The MCAT test-writers are skilled at presenting data in ways that can mislead a rushed reader. Watch for these tricks.

Truncated Y-Axis

The y-axis does not start at zero. A bar graph showing values of 98, 100, and 102 looks like a big difference if the y-axis runs from 97 to 103. In reality, the differences are tiny. Always check where the axis starts.

Misleading Scales

One axis uses a logarithmic scale while the other uses a linear scale. A straight line on a log-linear graph actually represents exponential growth, not linear growth. Check axis labels carefully.

Dual Y-Axes

Two different variables are plotted on the same graph, each with its own y-axis (one on the left, one on the right). This can create the illusion of a relationship between variables that are actually on completely different scales.

Unlabeled or Mislabeled Axes

Always confirm what each axis measures and what units are used. A graph showing drug concentration in mg/L tells a very different story than one showing concentration in g/L.

Imagine you are trying to describe the “typical” income in a neighborhood. You could average everyone’s salary, find the middle salary, or identify the most common salary. Each approach gives a different answer, and each tells a different story. Central tendency is about summarizing a dataset with a single representative number - but choosing the wrong measure can be seriously misleading.

Mean (Average)

The mean is the sum of all values divided by the number of values. It is the most commonly used measure of central tendency.

Mean = (sum of all values) / n

The mean uses every data point in the calculation, which makes it powerful but also makes it vulnerable. A single extreme value can drag the mean far from where most of the data actually lives.

Median

The median is the middle value when all data points are arranged in order. If there is an even number of values, the median is the average of the two middle values.

The median is resistant to outliers. No matter how extreme the highest or lowest value is, the median stays anchored in the center of the data. This makes it the preferred measure for skewed distributions.

Mode

The mode is the most frequently occurring value in a dataset. A dataset can have one mode (unimodal), two modes (bimodal), or more (multimodal). If no value repeats, there is no mode.

The mode is the only measure of central tendency that works for categorical data. You cannot calculate the mean or median of eye colors, but you can find the most common eye color (the mode).

Symmetric vs. Skewed Distributions

The relationship between mean, median, and mode tells you about the shape of a distribution.

Symmetric distribution: Mean = Median = Mode. The data is balanced around the center. A bell curve is the classic example.

Right-skewed (positive skew): Mean > Median > Mode. A long tail stretches to the right, pulling the mean in that direction. Example: household income in most countries.

Left-skewed (negative skew): Mean < Median < Mode. A long tail stretches to the left, pulling the mean down. Example: age at retirement (most people retire around 65, but a few retire very young).

Visualizing the Order

For a right-skewed distribution, picture the values along a number line from left to right:

Mode - Median - Mean (mean pulled right toward the tail)

For a left-skewed distribution:

Mean - Median - Mode (mean pulled left toward the tail)

When to Use Each Measure

Knowing the center of your data is only half the story. Two classes can both have a mean exam score of 75%, yet tell completely different stories. In one class, every student scored between 70% and 80%. In the other, scores ranged from 30% to 100%. The averages are identical, but the experiences are worlds apart. Variability - how spread out the data is - gives you the other half of the picture.

Range

The range is the simplest measure of spread: the difference between the maximum and minimum values.

Range = Maximum - Minimum

It is easy to calculate but very crude. A single outlier can blow up the range. If 99 students score between 70 and 80 and one student scores 15, the range jumps from 10 to 65 - even though almost all the data is tightly clustered.

Variance

Variance measures how far each data point is from the mean, on average. The calculation involves three steps:

1. Find the mean of the dataset.
2. Subtract the mean from each data point to get the deviation. Square each deviation.
3. Average the squared deviations.

Variance (s²) = Σ$(x - \bar{x})^2$ / (n - 1) for a sample

Why square the deviations? Because raw deviations (positive and negative) cancel each other out and always sum to zero. Squaring ensures all deviations are positive.

The catch: because we squared the deviations, the units of variance are squared too. If your data is in centimeters, the variance is in cm². That makes variance hard to interpret directly, which is why we usually take one more step.

Standard Deviation

Standard deviation (SD or s) is simply the square root of the variance. This brings the units back to match the original data.

Standard deviation = sqrt(variance)

What Standard Deviation Tells You

Think of SD as a “typical distance from the mean.” If the mean height in a class is 170 cm with SD = 5 cm, most students are within about 5 cm of 170 cm. If SD = 15 cm, heights are much more spread out.

Why n - 1?

You may notice the sample variance formula divides by (n - 1) instead of n. This is called Bessel’s correction. A sample tends to underestimate the true population variance because sample data points cluster closer to the sample mean than to the true population mean. Dividing by (n - 1) slightly inflates the estimate to correct for this bias. For the MCAT, just know that sample variance uses (n - 1) and population variance uses N.

Interquartile Range (IQR)

The interquartile range is the range of the middle 50% of the data. It is calculated by finding the first quartile (Q1, the 25th percentile) and third quartile (Q3, the 75th percentile), then subtracting.

IQR = Q3 - Q1

Like the median, the IQR is resistant to outliers. It ignores the top 25% and bottom 25% of the data entirely, focusing only on the central bulk. This makes it the go-to measure of spread for skewed distributions.

Comparing Measures of Variability

Your professor announces that the exam will be “graded on a curve.” What does that actually mean? It means the raw scores will be mapped onto a normal distribution - the famous bell-shaped curve where most students cluster around the average and fewer students appear at the extremes. This single shape turns up everywhere: heights of adults, blood pressure readings, measurement errors in a lab, IQ scores. Understanding its properties is essential for nearly every statistics question on the MCAT.

What Makes a Distribution “Normal”?

A normal distribution has several defining features:

• Bell-shaped and symmetric around the mean. The left half is a mirror image of the right half.
• Mean = Median = Mode. All three measures of central tendency sit at the center.
• Tails extend infinitely in both directions but never touch the x-axis. Extreme values are possible but increasingly rare.
• Completely defined by two numbers: the mean (which sets the center) and the standard deviation (which sets the width).

The 68-95-99.7 Rule (Empirical Rule)

This is the single most important fact about the normal distribution for the MCAT. Memorize it cold.

• 68% of data falls within 1 SD of the mean (between -1 SD and +1 SD)
• 95% of data falls within 2 SDs of the mean (between -2 SD and +2 SD)
• 99.7% of data falls within 3 SDs of the mean (between -3 SD and +3 SD)

Worked Example

Suppose IQ scores follow a normal distribution with mean = 100 and SD = 15.

• 68% of people have IQs between 85 and 115 (100 +/- 15)
• 95% of people have IQs between 70 and 130 (100 +/- 30)
• 99.7% of people have IQs between 55 and 145 (100 +/- 45)

An IQ of 130 is exactly 2 SDs above the mean. Since 95% of the data falls within 2 SDs, only 5% falls outside this range. Because the distribution is symmetric, 2.5% falls above 130 and 2.5% falls below 70.

Tail Percentages Quick Reference

Shifting and Stretching the Curve

Changing the mean shifts the entire bell curve left or right along the number line without changing its shape. Changing the SD changes the width: a larger SD makes the curve wider and flatter, while a smaller SD makes it taller and narrower. The total area under the curve is always 1 (representing 100% of the data), so a wider curve must be shorter to compensate.

You scored 720 on the biology section and 510 on the psychology section. Which performance was better? You cannot compare raw scores directly because the two sections have different means and different spreads. You need a common yardstick - a way to say “how far above or below average was this score, relative to everyone else?” That yardstick is the z-score.

The Z-Score Formula

A z-score tells you how many standard deviations a data point is above or below the mean.

z = (x - μ) / σ

where x is the individual value, μ is the population mean, and σ is the standard deviation.

Interpreting Z-Scores

• z = 0: The value equals the mean. Perfectly average.
• z = +1: The value is 1 SD above the mean.
• z = +2: The value is 2 SDs above the mean.
• z = -1: The value is 1 SD below the mean.
• z = -1.5: The value is 1.5 SDs below the mean.

Positive z-scores are above average. Negative z-scores are below average. The magnitude tells you how far from average.

Z-Scores to Percentiles

By combining z-scores with the 68-95-99.7 rule, you can quickly estimate percentiles.

Worked Example

A medical school class has a mean Step 1 score of 230 with SD = 20. A student scored 270.

z = (270 - 230) / 20 = 40 / 20 = 2

This student is exactly 2 SDs above the mean. From the table above, z = +2 corresponds to the 97.5th percentile. This student scored higher than 97.5% of the class.

Comparing Across Different Scales

Z-scores let you compare values from completely different distributions. Suppose:

• Biology exam: your score = 82, mean = 75, SD = 5. Your z = (82 - 75)/5 = 1.4
• Chemistry exam: your score = 68, mean = 60, SD = 4. Your z = (68 - 60)/4 = 2.0

Even though your raw biology score (82) was higher, your chemistry performance (z = 2.0) was relatively better compared to the class. You were further above average in chemistry.

You read that a new drug lowers blood pressure by 8 mmHg, with a 95% confidence interval of 5 to 11 mmHg. What does that actually tell you? Is it just a fancy way of saying “somewhere between 5 and 11”? Not exactly. Confidence intervals are one of the most commonly misinterpreted concepts in all of statistics, and the MCAT knows it. Getting the interpretation right - and understanding the common wrong interpretation - is a high-yield skill.

What Is a Confidence Interval?

A confidence interval (CI) is a range of values that is likely to contain the true population parameter. When researchers report a result, they are working with a sample, not the entire population. The CI acknowledges that the sample estimate is imprecise and provides a range.

The Correct Interpretation

“If we repeated this experiment many times, 95% of the resulting confidence intervals would contain the true population parameter.”

This is a statement about the procedure, not about any single interval. Once a CI is calculated (say, 5 to 11 mmHg), the true value either is or is not in that range - we just do not know which.

What Affects the Width of a CI?

A narrow CI means high precision. A wide CI means low precision. Three factors control the width:

1. Sample Size

Larger samples produce narrower CIs. This is the most important factor. Doubling the sample size does not halve the CI width (it shrinks by a factor of sqrt(2)), but increasing sample size always improves precision.

2. Variability (SD)

More variable data produces wider CIs. If individual measurements scatter widely, the estimate of the mean is less precise.

3. Confidence Level

A higher confidence level (e.g., 99% vs. 95%) produces a wider CI. Wanting to be more confident that you caught the true value requires casting a wider net. A 99% CI is always wider than a 95% CI from the same data.

CIs and Statistical Significance

Confidence intervals provide an elegant way to assess statistical significance without even looking at a p-value.

If a 95% CI for a difference between two groups does NOT include zero, the difference is statistically significant at p < 0.05.

If the 95% CI DOES include zero, the difference is NOT statistically significant at the 0.05 level.

Why? If the CI for a treatment effect includes zero, then “no effect” is a plausible value. You cannot rule out the possibility that the true effect is zero.

Overlapping CIs Between Groups

When comparing two groups, if their individual 95% CIs do not overlap at all, the difference is definitely significant. However, if the CIs do overlap slightly, the difference might still be significant - overlapping CIs is not the same as a non-significant difference. The proper test is to look at the CI of the difference, not the overlap of individual CIs.

A pharmaceutical company claims their new drug lowers cholesterol. A skeptic says it does nothing. How do you settle the debate with data? You cannot just eyeball a graph and declare a winner. You need a formal framework - a set of rules that everyone agrees on before the experiment starts - that tells you when the evidence is strong enough to side with the company over the skeptic. That framework is hypothesis testing, and it is the backbone of every research study you will encounter on the MCAT.

The Jury Trial Analogy

Hypothesis testing works exactly like a criminal trial.

The Null Hypothesis (H-null)

The null hypothesis ($H_0$) is the default assumption: there is no effect, no difference, no relationship. It represents the skeptic’s position.

Examples:

• “The drug has no effect on cholesterol.” ($H_0$: μ-drug = μ-placebo)
• “There is no difference in test scores between the two teaching methods.”
• “There is no correlation between sleep and GPA.”

The null hypothesis always contains an equals sign. It claims nothing interesting is happening.

The Alternative Hypothesis ($H_1$ or Hₐ)

The alternative hypothesis is the researcher’s claim: there IS an effect, a difference, or a relationship. It is what you are trying to find evidence for.

Examples:

• “The drug lowers cholesterol.” ($H_1$: μ-drug < μ-placebo)
• “The two teaching methods produce different test scores.”
• “There is a correlation between sleep and GPA.”

The Steps of Hypothesis Testing

1. State hypotheses. Define $H_0$ (no effect) and $H_1$ (there is an effect).
2. Set the significance level (α). By convention, α = 0.05. This is the threshold for “strong enough evidence.”
3. Collect data and calculate a test statistic. The test statistic measures how far the sample result is from what $H_0$ predicts.
4. Find the p-value. The probability of getting a result this extreme (or more extreme) if $H_0$ is true.
5. Make a decision. If p < α, reject $H_0$. If p >= α, fail to reject $H_0$.

”Reject” vs. “Fail to Reject”

This language is deliberate and the MCAT tests it.

Reject $H_0$: The evidence is strong enough to conclude that the null hypothesis is unlikely. The result is “statistically significant.”

Fail to reject $H_0$: The evidence is not strong enough to rule out the null hypothesis. This does NOT mean $H_0$ is true - it means you do not have enough evidence to reject it.

One-Tailed vs. Two-Tailed Tests

A two-tailed test checks for a difference in either direction. $H_1$: the drug changes cholesterol (could go up or down). The α is split between both tails (0.025 in each tail for α = 0.05).

A one-tailed test checks for a difference in a specific direction. $H_1$: the drug lowers cholesterol. All of α (0.05) is in one tail, making it easier to achieve significance in that direction.

The p-value might be the single most misunderstood number in all of science. Researchers misinterpret it. Journalists butcher it. Even some textbooks get it wrong. The MCAT knows this, and it will test whether you understand what a p-value actually means - and more importantly, what it does not mean. Get this concept right and you have a major advantage on test day.

The Definition

A p-value is the probability of obtaining results at least as extreme as the observed results, assuming the null hypothesis is true.

Read that again slowly. The p-value does not tell you the probability that the null hypothesis is true or false. It tells you how surprising your data would be in a world where the null hypothesis is true.

The 0.05 Threshold

By convention, a p-value less than 0.05 is considered “statistically significant.” This means:

• p < 0.05: The result is unlikely enough under $H_0$ that we reject the null hypothesis. “Statistically significant.”
• p >= 0.05: The result is not surprising enough under $H_0$ to reject it. “Not statistically significant.”

The 0.05 cutoff is a convention, not a law of nature. Some fields use stricter thresholds (particle physics uses p < 0.0000003). The MCAT almost always uses 0.05 unless stated otherwise.

What the P-Value is NOT

These misconceptions appear as trap answer choices on the MCAT.

Misconception 1: “The p-value is the probability that $H_0$ is true.”

Wrong. The p-value assumes $H_0$ is true and then asks how likely the data is. It does not give you the probability that $H_0$ is true or false.

Misconception 2: “p = 0.03 means there is a 3% chance the result is due to chance.”

Wrong. The p-value is the probability of seeing data this extreme if $H_0$ is true, not the probability that the result is a fluke.

Misconception 3: “A smaller p-value means a larger effect.”

Wrong. A small p-value means the evidence against $H_0$ is strong, but it says nothing about the size of the effect. You can get a tiny p-value from a huge sample even if the actual effect is trivially small.

P-Values and Sample Size

This is a critical nuance the MCAT tests. A large sample size increases the power of a study, making it easier to detect small effects. This means:

• A study with 100,000 participants can produce a tiny p-value (say, p = 0.0001) for a completely trivial effect (like a drug that lowers blood pressure by 0.1 mmHg).
• A study with 15 participants might produce a large p-value (say, p = 0.15) even if the drug works, simply because the sample was too small to detect the effect reliably.

P-Values in MCAT Passages

Research passages will often state something like: “The difference between groups was significant (p = 0.03)” or “No significant difference was found (p = 0.42).” When you see this, you should be able to:

1. Determine whether $H_0$ was rejected (p < 0.05) or not (p >= 0.05).
2. Understand that this is not the probability that the null is true.
3. Consider whether the sample size might be influencing the result.

Every time you make a decision based on data, there are exactly two ways to be wrong. You can sound the alarm when there is no danger, or you can stay silent when the building is on fire. In statistics, these are called Type I and Type II errors, and the MCAT absolutely loves testing them. The good news: once you learn the fire alarm analogy, you will never confuse them again.

Type I Error (False Positive)

A Type I error occurs when you reject the null hypothesis even though it is actually true. You concluded there was an effect when there really was not one.

Type I = False Positive = “Seeing something that is not there”

The probability of a Type I error is α, which is the significance level you set before the experiment. If α = 0.05, you accept a 5% risk of a Type I error.

Type II Error (False Negative)

A Type II error occurs when you fail to reject the null hypothesis even though it is actually false. You missed a real effect.

Type II = False Negative = “Missing something that IS there”

The probability of a Type II error is β. Unlike α, β is not directly chosen by the researcher - it depends on sample size, effect size, and variability.

The Error Table

Statistical Power

Power is the probability of correctly rejecting a false null hypothesis - in other words, the probability of detecting a real effect when one exists.

Power = 1 - β

If β (probability of a Type II error) is 0.20, then power = 0.80, meaning there is an 80% chance of detecting a real effect. Most well-designed studies aim for power of at least 0.80.

How to Increase Power

There are four main ways to increase the power of a study:

1. Increase sample size. More participants means more data, which makes it easier to detect real effects. This is the most common and practical approach.
2. Increase the effect size. A larger true effect is easier to detect. Researchers cannot usually control this, but choosing a more potent treatment helps.
3. Decrease variability. Tighter controls and more precise measurements reduce noise, making the signal easier to see.
4. Increase α. Using α = 0.10 instead of 0.05 makes it easier to reject $H_0$, which increases power but also increases the Type I error rate.

The Trade-Off

Reducing one type of error increases the other, all else being equal.

• If you lower α (stricter threshold, e.g., from 0.05 to 0.01), you reduce Type I errors but increase Type II errors. You are less likely to sound a false alarm, but more likely to miss a real fire.
• If you raise α (more lenient, e.g., from 0.05 to 0.10), you reduce Type II errors but increase Type I errors. You catch more real effects but also get more false alarms.

The only way to reduce both error types simultaneously is to increase sample size or reduce variability.

Ice cream sales and drowning deaths are strongly correlated. As ice cream sales go up, drowning deaths go up. Does ice cream cause drowning? Obviously not. Both increase during summer because of hot weather - a third variable that drives both. This is the most important lesson in all of statistics, and it appears on the MCAT more than almost any other statistical concept: correlation does not equal causation.

The Correlation Coefficient (r)

The correlation coefficient r quantifies the strength and direction of a linear relationship between two variables. It ranges from -1 to +1.

• r = +1: Perfect positive correlation. As x increases, y increases in a perfectly linear pattern.
• r = -1: Perfect negative correlation. As x increases, y decreases in a perfectly linear pattern.
• r = 0: No linear relationship. Knowing x tells you nothing about y.

Values between these extremes indicate the strength of the correlation:

The same ranges apply to negative correlations (just add a minus sign).

What r Does NOT Tell You

1. Causation

A strong correlation (even r = 0.95) does not mean x causes y. It only means they move together. They might both be caused by a third variable.

2. Nonlinear Relationships

r measures only linear relationships. Two variables could have a perfect curved relationship (like a parabola) and still have r = 0. If a scatter plot shows a clear U-shape, the correlation coefficient will underestimate the true relationship.

3. Appropriateness for Outliers

A single extreme outlier can sharply inflate or deflate r. Always look at the scatter plot, not just the number.

Correlation vs. Causation: The Three Criteria

For a researcher to claim that A causes B, three conditions must be met:

1. Correlation exists. A and B must be associated (necessary but not sufficient).
2. Temporal precedence. A must come before B in time. The cause must precede the effect.
3. Elimination of alternative explanations. Other variables (confounders) that could explain the relationship must be ruled out. This is why randomized controlled trials are the gold standard - randomization distributes confounders equally between groups.

Confounding Variables (Third Variables)

A confounding variable is a hidden factor that is related to both the independent and dependent variables, creating the illusion of a direct relationship between them.

Classic examples:

• Ice cream sales and drowning: Confound = summer heat (causes both).
• Shoe size and reading ability in children: Confound = age (older children have bigger feet AND read better).
• Number of firefighters at a fire and property damage: Confound = size of fire (bigger fires bring more firefighters AND cause more damage).

Study Types and Causal Claims

Here is the good news: the MCAT will never ask you to calculate a t-test statistic, run an ANOVA, or perform a chi-square test by hand. But it will absolutely show you a research passage and ask which test the researchers should have used - or whether the test they chose was appropriate. Think of it like knowing which tool to grab from a toolbox without needing to build the tool yourself.

The Decision: What Kind of Data Do You Have?

Choosing the right statistical test comes down to two questions:

1. What type of data is the outcome? Continuous (numbers on a scale, like blood pressure) or categorical (groups, like “survived” vs. “died”)?
2. How many groups are being compared?

T-Test: Comparing Means of Two Groups

Use a t-test when you are comparing the average (mean) of a continuous variable between exactly two groups.

Examples:

• Does Drug A lower blood pressure more than placebo? (Drug group mean vs. placebo group mean)
• Do men and women differ in average resting heart rate?
• Is there a difference in exam scores before vs. after a tutoring program? (paired t-test)

There are two main variants:

• Independent t-test: Two separate groups (e.g., treatment vs. control).
• Paired t-test: The same group measured twice (e.g., before and after treatment).

ANOVA: Comparing Means of Three or More Groups

ANOVA (Analysis of Variance) extends the logic of the t-test to three or more groups. Instead of asking “do these two groups differ?” it asks “do any of these groups differ from any other?”

Examples:

• Comparing average recovery time across three treatment protocols.
• Testing whether four different diets produce different weight loss.
• Comparing exam scores across five study methods.

ANOVA produces a single p-value that tells you whether at least one group differs from the others. It does not tell you which specific groups differ - that requires follow-up tests (post hoc tests).

Chi-Square Test: Comparing Frequencies (Categorical Data)

Use a chi-square test when both your variables are categorical and you are comparing observed frequencies to expected frequencies.

Examples:

• Does the ratio of blood types in a sample differ from the expected population ratio?
• Is there an association between smoking status (yes/no) and lung cancer (yes/no)?
• Does a coin-flip experiment match the expected $\frac{50}{50}$ distribution?

Chi-square is the test to use when your data is counts in categories, not measurements on a continuous scale.

Correlation and Regression: Relationships Between Continuous Variables

When you want to measure the relationship between two continuous variables (not compare groups), use correlation or regression.

• Correlation gives you the r value: the strength and direction of the linear relationship.
• Linear regression gives you the equation of the best-fit line (y = mx + b), allowing prediction. It tells you how much y changes for each unit change in x.

Examples:

• What is the relationship between study hours and exam scores?
• Does increasing drug dose predict blood pressure reduction?
• Is there a linear relationship between age and reaction time?

The Decision Table

A new blood pressure drug is tested in 200,000 patients. It lowers systolic blood pressure by 0.5 mmHg compared to placebo, with p = 0.001. The result is statistically significant. Should doctors prescribe it? Absolutely not. A 0.5 mmHg reduction is clinically meaningless - you would not even notice it on a blood pressure cuff. This study perfectly illustrates the most important concept in this entire chapter: statistical significance does not mean the result matters.

Statistical Significance vs. Clinical Significance

Statistical significance (p < 0.05) tells you the result is unlikely to be due to chance alone. It says nothing about the size or importance of the effect.

Clinical significance (or practical significance) asks: is the effect large enough to actually matter in the real world?

Effect Size

Effect size quantifies the magnitude of a difference, independent of sample size. While p-values shrink as sample size grows (even for tiny effects), effect size stays constant.

Cohen’s d

Cohen’s d is the most commonly referenced effect size measure. It expresses the difference between two group means in units of standard deviations.

d = (Mea$n_1$ - Mea$n_2$) / pooled SD

Cohen’s benchmarks:

• d = 0.2: Small effect
• d = 0.5: Medium effect
• d = 0.8: Large effect

A drug that lowers blood pressure by 0.5 mmHg when the SD of blood pressure is 15 mmHg has d = $\frac{0.5}{15}$ = 0.03 - a negligibly small effect, no matter how small the p-value.

A therapy that improves depression scores by 8 points on a scale with SD = 10 has d = 0.8 - a large, clinically meaningful effect.

SD vs. SEM: Two Very Different Error Measures

This distinction is tested directly on the MCAT and is crucial for interpreting graphs with error bars.

Standard Deviation (SD)

SD measures the spread of individual data points around the mean. It tells you how variable the raw data is.

SD answers: “How spread out are individual measurements?”

Standard Error of the Mean (SEM)

SEM measures the precision of the sample mean as an estimate of the population mean. It is calculated as:

SEM = SD / sqrt(n)

SEM is always smaller than SD (because you divide by sqrt(n)). As sample size increases, SEM decreases, meaning the estimate of the mean becomes more precise.

SEM answers: “How precisely do we know the true mean?”

Error Bars: Check the Label

Error bars on graphs can represent SD, SEM, or 95% confidence intervals. The interpretation changes a lot depending on which is used.

• SD error bars: Show the spread of the data. About 68% of data points fall within one SD bar above and below the mean.
• SEM error bars: Show the precision of the mean estimate. They are much smaller than SD bars and can make differences look more impressive.
• 95% CI error bars: If the error bars of two groups do not overlap, the difference is statistically significant. (Note: slight overlap does not necessarily mean non-significance.)

Putting It All Together

When you encounter a research result on the MCAT, evaluate it on three dimensions:

1. Statistical significance: Is p < 0.05? If not, the result may be due to chance.
2. Effect size: Is the difference large enough to matter? Use Cohen’s d benchmarks.
3. Precision: How wide is the confidence interval? Wide CIs suggest the true effect could be much larger or smaller than reported.

A well-designed, high-quality study has all three: a significant p-value, a meaningful effect size, and a narrow confidence interval.