# Fundamental Statistics: Core Concepts Explained

Fundamental statistics is the set of methods used to summarize data and to judge how much of a pattern in that data reflects a real effect rather than random variation. Descriptive statistics compress a dataset into a few numbers (mean, median, variance, standard deviation), while inferential statistics use those numbers to test hypotheses and estimate confidence intervals.

Every biological measurement carries noise. A dog's body weight shifts with hydration, time of day, and the scale itself. A cell count varies between replicate wells. Fundamental statistics gives you a disciplined way to separate signal from that noise, and it gives reviewers a shared language to check whether your conclusion survives scrutiny. Without it, a researcher cannot tell a genuine treatment effect from a lucky sample, and a clinician cannot tell whether a reported difference matters for a patient.

This guide walks through the core concepts in the order you actually use them: describing a dataset, then testing a claim about it. It uses a small veterinary dataset throughout so the arithmetic stays visible.

## Describing Data: Central Tendency and Spread

<figure class="article-figure">
  <img src="https://thumb.wikimedia.org/wikipedia/commons/thumb/7/74/Normal_Distribution_PDF.svg/1280px-Normal_Distribution_PDF.svg.png" alt="Several normal distribution curves with varying means and variances, labeled with μ and σ²" loading="lazy" decoding="async" width="1000" height="639" />
  <figcaption>Varying the mean shifts the curve while variance changes its spread, illustrating central tendency and spread. Image: Inductiveload, Public domain, via <a href="https://commons.wikimedia.org/wiki/File:Normal_Distribution_PDF.svg" rel="noopener noreferrer">Wikimedia Commons</a>.</figcaption>
</figure>

### The Mean

The mean is the arithmetic average. Add every value and divide by the number of values.

For a sample of n observations, the mean is written x̄ (x-bar) and calculated as the sum of all values divided by n. If five dogs weigh 10, 12, 14, 16, and 18 kg, the mean is (10+12+14+16+18)/5 = 70/5 = 14 kg.

The mean uses every data point, which makes it sensitive to extreme values. One 60 kg outlier in that group of five would pull the mean to 20 kg, far from where most dogs sit. The mean is the right summary when data are roughly symmetric and you care about the total. It is also the foundation of variance, standard deviation, and most hypothesis tests.

### The Median

The median is the middle value when observations are ranked from smallest to largest. With an odd number of values, it is the single central one. With an even number, it is the average of the two central values.

For the weights 10, 12, 14, 16, 60 kg, the median is 14 kg. The mean of the same data is 22.4 kg. The median barely moved because it depends on position, not magnitude. That resistance to outliers is why median income, median survival time, and median body condition score are reported so often. When a distribution is skewed, the mean and median diverge, and the gap between them is itself a useful clue about the shape of the data [1].

### Variance

Variance measures how far values spread around the mean. It is the average squared deviation from the mean.

The logic runs in three steps. First, subtract the mean from each value to get deviations. Second, square each deviation so positive and negative values do not cancel. Third, average those squared deviations. For a sample, the divisor is n minus 1 rather than n, a correction that makes the sample variance an unbiased estimate of the population variance.

Take four dog weights: 10, 12, 14, 16 kg, mean 13 kg. Deviations are -3, -1, +1, +3. Squared: 9, 1, 1, 9. Sum is 20. Divide by n minus 1 (3): variance = 6.67 kg².

Notice the units. Variance is in squared kilograms, which is why it is rarely reported as a headline number. It is a mathematical waypoint, not a directly interpretable quantity. Its value is that it feeds into standard deviation and into the analysis of variance used to compare group means [2].

### Standard Deviation

Standard deviation (SD) is the square root of the variance. It returns the spread to the original units.

For the four weights above, SD = √6.67 = 2.58 kg. That number is directly meaningful: it says a typical dog in this sample sits about 2.6 kg away from the mean.

SD describes the spread of the data. Roughly two-thirds of values in a roughly bell-shaped distribution fall within one SD of the mean, and about 95 percent fall within two SDs. Reporting "mean 13 kg, SD 2.6 kg" tells a reader both the center and how variable the individuals are. That is a statement about the animals, not about your estimate of the average [1].

### Standard Error of the Mean

Standard error of the mean (SEM) is the SD divided by the square root of n. It measures the precision of the mean as an estimate of the population mean.

With SD = 2.58 kg and n = 4, SEM = 2.58/2 = 1.29 kg. The units are still kilograms, but the meaning has changed. SD tells you how much individual dogs vary. SEM tells you how tightly your estimate of the average is pinned down. More samples shrink SEM, because the square root of n grows.

This distinction is one of the most common points of confusion in the life science literature. SD and SEM share units but answer different questions, and a small SEM can be produced by a large sample even when the underlying animals vary widely. When you read a figure with error bars, check the caption. Error bars labeled SD and error bars labeled SEM look identical but convey different information.

### Summary Table of Core Statistics

| Statistic | Formula (sample) | What it tells you | Units |
|--|--|--|--|
| Mean | Sum of values ÷ n | Center of the data | Same as data (kg) |
| Median | Middle ranked value | Center, resistant to outliers | Same as data (kg) |
| Variance | Sum of squared deviations ÷ (n - 1) | Average squared spread | Data units squared (kg²) |
| Standard deviation | Square root of variance | Typical distance from the mean | Same as data (kg) |
| Standard error of the mean | SD ÷ √n | Precision of the mean estimate | Same as data (kg) |
| 95% confidence interval | Mean ± (t × SEM) | Range of plausible population means | Same as data (kg) |
| p-value | Probability of data this extreme if H₀ is true | Compatibility of data with the null | Unitless (0 to 1) |

## From Description to Inference

### The Logic of Null-Hypothesis Significance Testing

Null-hypothesis significance testing (NHST) asks a single question: if there were truly no effect, how surprising would the data I collected be?

The procedure has four steps.

1. State a null hypothesis (H₀). This is usually "no difference" or "no association." For a diet study, H₀ says the mean weight before and after the diet is the same in the population.
2. State an alternative hypothesis (H₁). This is what you would conclude if the data are incompatible with H₀, for example that the mean weight changed.
3. Choose a test statistic and compute it from the data. A t-test compares two means relative to the variability within the groups.
4. Convert the test statistic into a p-value and compare it to a pre-set threshold, conventionally 0.05.

The p-value is the probability of observing a result at least as extreme as yours, assuming H₀ is true [3][4]. A small p-value means your data would be unusual under the null. It does not mean the null is false with that probability, and it does not measure the size of the effect [5].

### A Worked Example: Body Weight in 20 Dogs Before and After a Diet

Consider a small veterinary study. Twenty dogs are weighed before starting a weight-management diet and again after 12 weeks. For simplicity, treat the before and after measurements as paired observations on the same 20 animals.

Suppose the mean weight before the diet is 24.0 kg with SD 4.0 kg. After the diet, the mean is 22.8 kg with SD 4.1 kg. The mean change is a loss of 1.2 kg. The SD of the individual weight changes is 1.6 kg.

The paired t-test uses the mean change and the standard error of the change. The SEM of the change is SD ÷ √n = 1.6 ÷ √20 = 1.6 ÷ 4.47 = 0.358 kg.

The t statistic is the mean change divided by its SEM: 1.2 ÷ 0.358 = 3.35.

With 19 degrees of freedom (n - 1), a t value of 3.35 corresponds to a two-sided p-value of about 0.003. Since 0.003 is below 0.05, the result is conventionally called statistically significant. The data are unlikely if there were truly no average change.

Now read that result carefully. The p-value says the observed 1.2 kg loss would be rare under the null. It does not say the diet caused a 1.2 kg loss in every dog, and it does not say 1.2 kg is clinically important. A 1.2 kg loss in a 24 kg dog is a 5 percent reduction, which many clinicians would consider modest. The p-value and the clinical judgment are separate questions [6].

### Confidence Intervals

A confidence interval (CI) gives a range of plausible values for the population parameter. The 95 percent CI for the mean change is the mean change plus or minus the critical t value times the SEM. For 19 degrees of freedom, the critical t is about 2.09.

Lower bound: 1.2 - (2.09 × 0.358) = 1.2 - 0.748 = 0.45 kg.
Upper bound: 1.2 + (2.09 × 0.358) = 1.2 + 0.748 = 1.95 kg.

The 95 percent CI is 0.45 to 1.95 kg. Because it excludes zero, the result is significant at the 0.05 level, which is consistent with the p-value. But the interval adds information the p-value cannot: it shows the plausible range of the true average effect. A CI that runs from 0.45 to 1.95 kg tells a clinician that the real average loss is probably small, even if statistically detectable. Many methodologists argue that confidence intervals should be reported alongside or instead of p-values because they convey both the direction and the magnitude of an effect [7][8].

### Sample Size and the Width of the Interval

The CI width is driven by SEM, which is SD divided by √n. Small samples produce wide intervals.

If the diet study had only 5 dogs instead of 20, the SEM of the change would be 1.6 ÷ √5 = 0.716 kg. The critical t with 4 degrees of freedom is about 2.78. The 95 percent CI would then run from 1.2 - (2.78 × 0.716) = -0.79 kg to 1.2 + (2.78 × 0.716) = 3.19 kg. That interval includes zero, so the same 1.2 kg mean change would not be significant in a sample of five. The effect did not change. The precision did [4].

This is the single most important practical lesson in fundamental statistics: a non-significant result in a small study often means "we could not tell," not "there is no effect."

## Common Mistakes and Limitations

### p-Values Do Not Measure Effect Size

A p-value combines the size of an effect with the precision of the estimate and the sample size. A tiny effect measured in a huge sample can produce a very small p-value. A large effect measured in a tiny sample can produce a large p-value. The p-value answers "could this be noise?" not "does this matter?" [6][4]. Always report the effect size (mean difference, risk ratio, correlation coefficient) and its confidence interval next to the p-value.

### p-Values Are Not the Probability That the Null Is True

The most persistent misinterpretation is reading a p-value of 0.03 as "a 3 percent chance the null hypothesis is true." That is not what it means. The p-value is the probability of the data given the null, not the probability of the null given the data. These two quantities can differ enormously, and the gap widens when multiple tests are run or when the prior probability of a real effect is low [5].

### Multiple Comparisons Inflate False Positives

Every hypothesis test carries a risk of a false positive, conventionally 5 percent. Run 20 independent tests on data where nothing is happening, and the expected number of p-values below 0.05 is one. Run 100 tests and you expect about five.

This is why studies that test many outcomes, many subgroups, or many gene variants need a correction. Methods such as the Bonferroni adjustment or stepdown procedures like the Romano-Wolf method control the family-wise error rate, the probability of at least one false positive across the whole family of tests [9]. When a paper reports a single significant result from a large panel of comparisons without any correction, treat it with caution [5].

### Small Samples Widen Confidence Intervals

As shown above, cutting the sample from 20 to 5 turned a significant result into a non-significant one without changing the underlying effect. Small studies are not wrong, they are imprecise. The correct interpretation of a wide CI is that the data are compatible with a broad range of true effects, including zero. Reporting only the p-value hides this [7].

### Statistical Significance Is Not Clinical Significance

A statistically significant 1.2 kg weight loss in dogs may or may not translate into better joint health, longer lifespan, or improved quality of life. [Statistical tests](/blog/guides/statistical-tests-choosing-the-right-one-for-your-data) operate on numbers. Clinical decisions require judgment about what those numbers mean for an animal. The two are related but not identical, and conflating them is a frequent error in both reading and writing research [10][6].

### The Replication Problem

The widespread misuse of p-values has contributed to a replication crisis across the life sciences. Many published findings fail to reproduce, and hypothesis testing is often blamed. The problems are not with the mathematics of the p-value but with how it is used: selective reporting, multiple comparisons without correction, small samples, and treating a threshold as a verdict [10][11]. Careful reporting of effect sizes, confidence intervals, and full methods reduces but does not eliminate this risk.

## How These Concepts Fit Together in Practice

When you read or write a results section, the numbers should appear in a logical order.

1. Report the sample size and the descriptive statistics for each group: mean, SD, and n. These describe the animals or samples you studied.
2. State the hypothesis and the test used. Name the test, the test statistic, and the degrees of freedom.
3. Report the effect size with its confidence interval. This is the number a reader will use to judge practical importance.
4. Report the p-value. This tells the reader how compatible the data are with the null.
5. Interpret in plain language, separating the statistical finding from the clinical or biological conclusion.

A results sentence that follows this order might read: "Mean body weight fell from 24.0 kg (SD 4.0) to 22.8 kg (SD 4.1) in 20 dogs, a mean reduction of 1.2 kg (95% CI 0.45 to 1.95, p = 0.003)." Every number has a unit, a meaning, and a purpose. Nothing is left for the reader to guess.

## Quick Review

- The mean is the arithmetic average and is sensitive to outliers. The median is the middle ranked value and resists them.
- Variance is the average squared deviation from the mean, in squared units. Standard deviation is its square root, back in the original units.
- Standard deviation describes how much individuals vary. Standard error of the mean describes how precisely the mean is estimated. Both are in the same units as the data.
- A p-value is the probability of data at least as extreme as yours, assuming the null hypothesis is true. It is not the probability that the null is true.
- A 95 percent confidence interval gives the plausible range for the true effect. Small samples produce wide intervals.
- Multiple comparisons inflate false positives. Correction methods control the family-wise error rate.
- [Statistical significance](/blog/guides/statistical-significance) and clinical significance are different judgments. Report effect sizes and confidence intervals, not just p-values.

## Frequently Asked Questions

### What is the difference between standard deviation and standard error?

Standard deviation measures how spread out individual observations are around the mean. Standard error measures how precisely the sample mean estimates the population mean. Both are in the same units as the data, but they answer different questions.

### What does a p-value of 0.05 actually mean?

It means that if the null hypothesis were true, data at least as extreme as yours would occur about 5 percent of the time. It does not mean there is a 5 percent chance the null is true, and it does not measure the size of the effect.

### Why do small studies often fail to find significant results?

Small samples produce large standard errors, which widen confidence intervals and reduce [statistical power](/blog/guides/statistical-power-what-it-is-and-why-it-matters-in-research). A real effect can exist but remain undetected because the study lacks the precision to distinguish it from noise.

### Can a result be statistically significant but not clinically important?

Yes. Large samples can make tiny effects statistically significant. A statistically significant change may be too small to affect an animal's health, so effect size and clinical context matter alongside the p-value.

### What is multiple comparisons and why does it matter?

Multiple comparisons means running many hypothesis tests on the same dataset. Each test carries a false-positive risk, so the chance of at least one spurious significant result grows quickly. Correction methods adjust for this.

### Should I report confidence intervals instead of p-values?

Report both. Confidence intervals show the range of plausible effect sizes and their precision. p-values show compatibility with the null. Together they give a fuller picture than either alone.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What is the difference between standard deviation and standard error?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Standard deviation measures how spread out individual observations are around the mean. Standard error measures how precisely the sample mean estimates the population mean. Both are in the same units as the data, but they answer different questions."
      }
    },
    {
      "@type": "Question",
      "name": "What does a p-value of 0.05 actually mean?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "It means that if the null hypothesis were true, data at least as extreme as yours would occur about 5 percent of the time. It does not mean there is a 5 percent chance the null is true, and it does not measure the size of the effect."
      }
    },
    {
      "@type": "Question",
      "name": "Why do small studies often fail to find significant results?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Small samples produce large standard errors, which widen confidence intervals and reduce statistical power. A real effect can exist but remain undetected because the study lacks the precision to distinguish it from noise."
      }
    },
    {
      "@type": "Question",
      "name": "Can a result be statistically significant but not clinically important?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes. Large samples can make tiny effects statistically significant. A statistically significant change may be too small to affect an animal's health, so effect size and clinical context matter alongside the p-value."
      }
    },
    {
      "@type": "Question",
      "name": "What is multiple comparisons and why does it matter?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Multiple comparisons means running many hypothesis tests on the same dataset. Each test carries a false-positive risk, so the chance of at least one spurious significant result grows quickly. Correction methods adjust for this."
      }
    },
    {
      "@type": "Question",
      "name": "Should I report confidence intervals instead of p-values?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Report both. Confidence intervals show the range of plausible effect sizes and their precision. p-values show compatibility with the null. Together they give a fuller picture than either alone."
      }
    }
  ]
}
</script>

## Related Articles

- [Core concepts scientific paper](/blog/research-skills/core-concepts-first-how-to-identify-and-focus-on-the-key-ideas-in-a-paper-from-a-different-disciplin)
- [Epigenetics Textbook: Core Concepts and Mechanisms Explained](/knowledge/molecular-biology/epigenetics-textbook)
- [Veterinary Pharmacology and Therapeutics: Core Concepts for Clinicians](/knowledge/veterinary-medicine/clinical-pharmacology/veterinary-pharmacology-therapeutics-core-concepts-clinicians)
- [Defining Ethics in Research: Core Concepts and Misconceptions](/blog/guides/defining-ethics-in-research-core-concepts-and-misconceptions)
- [Veterinary Public Health and Epidemiology: Core Concepts and Applications](/knowledge/veterinary-medicine/veterinary-public-health/veterinary-public-health-epidemiology-core-concepts-applications)
- [What Is Master Data Management? Core Concepts and Use Cases](/blog/guides/what-is-master-data-management-core-concepts-and-use-cases)
- [MAP and TAU in Bayesian Statistics](/blog/research-skills/map-and-tau-in-bayesian-statistics)
- [Statistical Difference: How to Tell If Results Differ](/blog/research-skills/statistical-difference-how-to-tell-if-results-differ)

## Sources

1. [Descriptive statistics: Measures of central tendency, dispersion, correlation and regression](https://doi.org/10.4103/ARWY.ARWY_37_19)
2. [On teaching about descriptive statistics in criminal justice: Mean, variance, and standard deviation](https://doi.org/10.1080/10511259900084661)
3. [Statistics review 3: Hypothesis testing and P values](https://pubmed.ncbi.nlm.nih.gov/12133182/)
4. [Decision Rules in Frequentist and Bayesian Hypothesis Testing: P-Value and Bayes Factor](https://pubmed.ncbi.nlm.nih.gov/40453986/)
5. [Misinterpreting p: The discrepancy between p values and the probability the null hypothesis is true, the influence of multiple testing, and implications for the replication crisis.](https://pubmed.ncbi.nlm.nih.gov/31829657/)
6. [Null Hypothesis Significance Testing, p-values, Effects Sizes and Confidence Intervals](https://doi.org/10.1017/BrImp.2017.28)
7. [Confidence intervals rather than P values: estimation rather than hypothesis testing.](https://pubmed.ncbi.nlm.nih.gov/3082422/)
8. [Hypothesis Testing with Confidence Intervals and P Values in PLS-SEM](https://doi.org/10.4018/IJeC.2016070101)
9. [RWOLF: Stata module to calculate Romano-Wolf stepdown p-values for multiple hypothesis testing](https://www.semanticscholar.org/paper/850f6ae62199b0cf15528d680dc33b59772f7f64)
10. [Editorial: Roles of Hypothesis Testing, p-Values and Decision Making in Biopharmaceutical Research](https://doi.org/10.1080/19466315.2021.1874803)
11. [P values, hypothesis testing, and model selection: it's déjà vu all over again.](https://pubmed.ncbi.nlm.nih.gov/24804440/)