Statistics for Researchers: A Complete Guide

Most researchers don't need to become statisticians. They need to choose the right analysis, run it correctly, interpret it honestly, and report it the way their field expects. This guide is the map for that work: statistics for researchers, organized in the order a real study moves through them.

Everything here links to a dedicated guide that works the topic out in full, with formulas, worked examples, and reporting conventions. Use this page to find where you are and what comes next, then follow the link that matches the decision in front of you.

Which Test Do You Need?

The most common question in applied statistics is which test to run. It comes down to what you're comparing and what kind of data you have.

The Shape of the Field

Statistics divides cleanly in two, and knowing which half you're in resolves most confusion about what a number is allowed to say.

Descriptive statistics describe the data in front of you. A mean, a standard deviation, a histogram: each one summarizes your sample and claims nothing beyond it. If you surveyed 200 students and found men reporting higher risk tolerance, descriptive statistics establish that this was true of those 200.

Inferential statistics take the next step. They use the sample to make a claim about a population you couldn't measure, and they quantify how uncertain that claim is. That second part is what makes inference more than guessing. A p-value, a confidence interval, and a power calculation are all ways of stating how much the sample could be misleading you.

Almost every quantitative paper uses both, in that order. You describe before you infer, because the shape and spread of your sample determine which inferential tools are valid on it.

Foundations: Who You're Studying and What You're Measuring

Before any analysis, two questions need clear answers. Who is this study about, and what exactly are you measuring?

The first is the distinction between the group you want to draw conclusions about and the smaller group you actually measure. Getting that relationship right is what licenses any claim beyond your own data. See the guide to population vs sample in research. The companion distinction is notational: a number describing a population is a parameter, one describing a sample is a statistic, and they use different symbols. The guide to parameter vs statistic covers the notation and where it trips people up.

The second question is about variables. Which factor are you treating as the cause, and which as the outcome? For what a predictor is and how it functions in a design, see the guide to independent variables. For the practical tests that tell the two roles apart in your own study, see identifying independent and dependent variables.

Sampling: Choosing Who to Study

How you select participants determines what your results can support. A biased selection method can't be rescued by a large sample, so this decision matters more than most researchers expect.

Methods that give every member of the population a known chance of selection are the ones that support formal inference. The overview of probability sampling covers the family. The four methods within it each suit a different situation:

• Simple random sampling, the baseline, where every member has an equal chance.
• Stratified sampling, when subgroups need guaranteed representation.
• Cluster sampling, when the population comes in natural groups and reaching individuals is impractical.
• Systematic sampling, when you select at a fixed interval from an ordered list.

Once the method is settled, the question is how many. That's a power analysis, and it's the one calculation that has to happen before data collection rather than after. See the guide to calculating sample size, which covers both the quantitative power analysis and the saturation logic qualitative studies use instead.

Describing Your Data

Every results section starts here. Descriptive statistics tell readers who was in your sample and what their values looked like, and they come before any test.

Three questions organize them. What's the typical value, how spread out is the data, and what shape is the distribution? For central tendency, see mean, median, and mode, which covers when each measure is the right choice. For spread, see standard deviation and variance. For shape, see skewness and kurtosis, which determines whether the mean or the median describes your data honestly.

Numbers alone compress a lot of information, so visualization belongs at this stage too. The guide to frequency distributions and histograms covers how to see a distribution's shape directly. For everything else, box plots, scatter plots, and choosing the right visualization gives a decision framework. The full treatment of this half of the field is in the descriptive statistics guide.

Distributions and Standardized Scores

One distribution underlies most of inferential statistics. Its properties are what let a test attach a probability to a result, which is why it's worth understanding before the tests themselves. See the guide to the normal distribution.

Standardized scores follow directly from it. Expressing a value as its distance from the mean in standard deviation units makes results from different scales directly comparable. The same standardization feeds into nearly every test statistic. See the guide to z-scores.

The Logic of Inference

Inference is the move from your sample to a claim about a population. Every specific test is a variation on one underlying procedure, so learning the procedure once makes the rest much easier.

That procedure is hypothesis testing. You state a null hypothesis of no effect, assume it's true, and ask how surprising your data would be under that assumption. See the guide to hypothesis testing and setting up the null and alternative.

The number that procedure produces is the p-value, and it's the most misread quantity in research. It's the probability of data at least as extreme as yours if the null were true. It is not the probability the null is true. See p-values explained for the misinterpretations that draw reviewer comments.

Every test can be wrong in two directions. A false positive reports an effect that isn't there. A false negative misses one that is. The guide to Type I and Type II errors covers both, along with the power that governs the second.

Testing isn't the only form of inference. Estimation gives a range of plausible values rather than a yes-or-no verdict, and many journals now prefer it. See confidence intervals. The full framework for this half of the field is in the inferential statistics guide.

Write the Analysis Plan Before You Collect Data

The single habit that prevents the most problems is deciding your analysis in advance and writing it down. Reviewers increasingly ask for it, and some journals require pre-registration.

A usable plan answers five questions. What are the hypotheses, stated as null and alternative? Which variables are predictors and which are outcomes? Which test will you run, and why does it suit your design and data type? What alpha level and power are you assuming, and what sample size do they require? And what will you do if an assumption fails?

Answering these in advance closes off the choices that damage credibility when made afterward. Consider picking a test once you've seen which one gives a significant result. Or switching from two-tailed to one-tailed to clear a threshold. Or deciding on subgroups after the fact. All three inflate your false positive rate. None of them look like misconduct while you're doing them, which is exactly why the plan matters.

The plan also makes the write-up faster. When the methodology section states the test, the inputs, and the reasoning before data collection, the results section becomes a matter of filling in numbers.

The Tests Themselves

With the framework in place, each test becomes a matter of matching the analysis to the question and the data.

Comparing groups

For one or two groups on a continuous outcome, see t-tests, which covers the one-sample, independent, and paired versions. For three or more groups, running repeated t-tests inflates your error rate, so see ANOVA instead. For categorical data, where you're comparing counts rather than means, see chi-square tests.

Measuring relationships

Correlation measures how strongly two continuous variables move together. For what the coefficient actually measures and how to read it, see correlation coefficients explained. For choosing between Pearson and Spearman and reporting the result in a manuscript, see using correlation coefficients in research papers.

Predicting outcomes

Regression extends correlation into prediction. Start with simple linear regression for one predictor, which introduces the slope, intercept, residuals, and R squared. Move to multiple regression when several predictors act at once, which is the usual case in real research. When the outcome is binary rather than continuous, see logistic regression.

When assumptions fail

The tests above assume things about your data, including normality and equal variances. When those assumptions don't hold, rank-based alternatives make fewer demands and remain valid. See when to use non-parametric tests.

Reporting Your Results

A correct analysis reported badly still gets sent back. Three things matter most here.

First, magnitude. Significance says an effect is detectable, not that it matters, and most journals now require an effect size alongside every test. See the guide to effect sizes for which measure pairs with which analysis.

Second, notation. Italics, leading zeros, degrees of freedom, and decimal conventions are things reviewers know by heart, and deviations get noticed immediately. See reporting statistical results in APA format.

Third, the errors themselves. Most statistical problems in published work are errors of interpretation and reporting rather than calculation, and they repeat predictably. See common statistics mistakes in academic papers for the full list organized by stage.

One practical matter often gets overlooked. Statistical notation in a manuscript has to be typeset correctly, and equation formatting is its own small skill. See how to write equations in Word.

One Question, All the Way Through

To see how the pieces connect, follow a single research question through the whole sequence. Fisher and Yao (2017) studied gender differences in financial risk tolerance using the Survey of Consumer Finances.

Every stage in that sequence has a guide above. The order rarely changes, and skipping a stage is where most problems start.

Getting Your Statistics Reviewed

Statistical writing is where a second reader adds the most value, because the errors that matter are errors of claim rather than calculation. Software returns the right number. What goes wrong is what the researcher says it means.

Editor World's academic editing, dissertation editing, and journal article editing services include review of statistical reporting. You choose your own editor by field, so the person reading your results knows your discipline's conventions. A certificate of editing confirming human-only native English editing is available as an optional add-on for journals that require an AI-use disclosure.

Frequently Asked Questions

Page last reviewed: September 2026. Content reviewed by Editor World editorial staff. Editor World, founded in 2010 by Patti Fisher, PhD, graduate of The Ohio State University, provides professional editing and proofreading services for academic researchers, doctoral candidates, faculty, business professionals, and authors worldwide. 100% human editing, no AI at any stage. BBB A+ accredited since 2010 with 5.0/5 Google Reviews and 5.0/5 Facebook Reviews. 16 years in business with 140 million+ words edited for over 8,000 clients in 65+ countries. Multiple Gold and Bronze Stevie Award winner. Native English editors from the United States, the United Kingdom, and Canada. Less than 5% of applicants are accepted to the editor panel. Recommended by the Boston University Economics Department, University of San Diego, University of Michigan, UCLA, University of Missouri, and more.