You have a dataset and a question. Maybe you need to know whether two groups are meaningfully different, or whether there is a relationship between advertising spend and revenue, or simply what the average and spread look like before presenting results to a stakeholder. In theory, you could fire up R, install packages, write scripts, and generate output. In practice, you just need the answer in the next five minutes.
Statistical analysis does not always require specialized software. The fundamentals, descriptive statistics, hypothesis tests, and basic regression, can be run in a browser when your dataset is small to medium and the question is straightforward. This guide covers the core statistical concepts you are most likely to need, walks through real examples with actual numbers, and shows when each analysis type is appropriate.
Photo by www.kaboompics.com on Pexels
Descriptive Statistics: Understanding Your Data
Before running any test, you need to know what your data looks like. Descriptive statistics summarize the center, spread, and shape of a dataset.
Mean, Median, and Mode
The mean (average) is the sum of all values divided by the count. It is sensitive to outliers. If ten employees earn between $45,000 and $65,000 but one earns $350,000, the mean salary of $72,727 misrepresents the typical worker.
The median is the middle value when data is sorted. It is robust to outliers. The median salary for the same group is $57,500, which better represents what most employees actually earn.
The mode is the most frequent value. It is useful for categorical data (the most common shirt size sold, the most popular response on a survey) but rarely informative for continuous data.
When to use each: Use the mean when your data is roughly symmetric with no extreme outliers. Use the median when outliers exist or the distribution is skewed. Use the mode for categorical data.
Standard Deviation and Variance
Standard deviation measures how spread out your data is around the mean. A small standard deviation means values cluster tightly. A large one means they are dispersed.
Take two classes of test scores: - Class A: 78, 80, 82, 79, 81 (mean = 80, SD = 1.58) - Class B: 60, 70, 80, 90, 100 (mean = 80, SD = 15.81)
Both classes have the same mean, but Class B has dramatically more variation. The standard deviation tells you that Class A performs consistently while Class B has a mix of struggling and excelling students. According to Khan Academy's statistics curriculum, understanding spread is often more actionable than understanding the center.
Variance is simply the standard deviation squared. It is used internally in many statistical formulas but is less interpretable than SD because its units are squared (dollars squared, points squared, etc.).
Percentiles and Quartiles
Percentiles tell you where a value falls relative to the rest. The 75th percentile means 75% of values are below this point. Quartiles split data into four equal parts: Q1 (25th percentile), Q2 (median, 50th), Q3 (75th). The interquartile range (IQR = Q3 - Q1) is another robust measure of spread that ignores outliers.
Photo by Karolina Grabowska www.kaboompics.com on Pexels
Hypothesis Testing: Answering "Is This Difference Real?"
Descriptive statistics tell you what happened. Hypothesis testing tells you whether what happened is statistically meaningful or could have occurred by chance.
The Logic of a Hypothesis Test
Every hypothesis test follows the same structure:
- Null hypothesis (H0): There is no effect or no difference. Any observed difference is due to random variation.
- Alternative hypothesis (H1): There is a real effect or difference.
- Test statistic: A number calculated from your data that measures how far the observed result is from what the null hypothesis predicts.
- P-value: The probability of observing a result at least as extreme as yours if the null hypothesis were true.
- Decision: If the p-value is below your significance threshold (typically 0.05), reject the null hypothesis.
A p-value of 0.03 does not mean there is a 3% chance the null hypothesis is true. It means that if the null hypothesis were true, you would see data this extreme only 3% of the time. The distinction matters. The American Statistical Association's statement on p-values clarifies this frequently misunderstood concept.
One-Sample t-Test
Use this when you want to test whether a sample mean differs from a known or hypothesized value.
Example: A coffee shop claims their medium coffee contains 16 ounces. You measure 10 cups and get: 15.2, 15.8, 15.5, 16.1, 15.3, 15.7, 15.9, 15.4, 16.0, 15.6. The sample mean is 15.65 ounces. Is this significantly less than 16?
Running a one-sample t-test: t = -3.72, p = 0.005. Since p < 0.05, you reject the null hypothesis. The coffee shop is statistically under-pouring.
Two-Sample t-Test
Use this when comparing the means of two independent groups.
Example: You run an A/B test on two landing page designs. Version A gets 50 visitors with a mean time-on-page of 45 seconds (SD = 12). Version B gets 50 visitors with a mean of 52 seconds (SD = 14). Is Version B actually better, or is this random variation?
Two-sample t-test: t = -2.69, p = 0.008. The difference is statistically significant. Version B genuinely keeps visitors on the page longer.
Z-Test for Proportions
Use this when comparing proportions rather than means.
Example: Your email campaign had a 4.2% click rate (42 out of 1,000) last month and 5.1% (51 out of 1,000) this month after changing the subject line. Is the improvement real?
Z-test for proportions: z = 0.96, p = 0.34. Not significant. The improvement is within the range of normal variation. You cannot conclude the new subject line is better.
The Statistics Calculator on EvvyTools runs all three test types. Paste your data, select the test, and it returns the test statistic, p-value, and a plain-language interpretation.
Photo by Negative Space on Pexels
Regression: Finding Relationships in Data
Regression analysis answers the question: is there a predictable relationship between two variables, and how strong is it?
Linear Regression in Practice
Example: You want to know if monthly ad spend predicts revenue. Here is data from 8 months:
Ad Spend ($): 2000, 3000, 4000, 5000, 6000, 7000, 8000, 9000 Revenue ($): 15000, 18000, 22000, 24000, 28000, 30000, 33000, 36000
Running linear regression produces: - Equation: Revenue = 2.93 x Ad Spend + 9,286 - R-squared: 0.99 - Interpretation: For every additional dollar of ad spend, revenue increases by $2.93. The model explains 99% of the variation in revenue.
An R-squared of 0.99 is unusually high. Real-world data is noisier. An R-squared of 0.5 to 0.7 is common and still useful for directional decision-making. The NIST Engineering Statistics Handbook provides guidelines for interpreting regression output in applied settings.
When Regression Misleads
Correlation does not imply causation. Your regression might show a strong relationship between ice cream sales and drowning deaths (both increase in summer), but ice cream does not cause drowning. Always consider confounding variables and the mechanism behind any relationship you find.
Also watch for extrapolation. If your ad spend data ranges from $2,000 to $9,000, your regression model has no basis for predicting what happens at $50,000. The linear relationship may not hold outside the observed range.
Quadratic Regression
Not all relationships are linear. Sometimes the data curves. Quadratic regression fits a parabola (y = ax^2 + bx + c) instead of a straight line. This is useful when you suspect diminishing returns (each additional dollar of ad spend produces slightly less revenue than the last) or accelerating growth (network effects, viral content).
The Statistics Calculator on EvvyTools supports both linear and quadratic regression. When in doubt, run both and compare the R-squared values. If quadratic produces a meaningfully higher R-squared (say, 0.85 vs 0.72), the curved model is a better fit for your data.
Probability Distributions
Beyond descriptive stats and hypothesis tests, the calculator handles three probability distributions that cover most practical needs:
Normal distribution: The bell curve. Given a mean and standard deviation, calculate the probability of a value falling within any range. Useful for quality control, grading curves, and any process with symmetric random variation.
Binomial distribution: The coin-flip model. Given a number of trials and a probability of success per trial, calculate the likelihood of exactly k successes. Useful for conversion rate analysis, A/B testing, and defect rate modeling.
Poisson distribution: The rare-event model. Given an average event rate, calculate the probability of seeing k events in a time period. Useful for website traffic analysis, call center staffing, and insurance claim modeling.
Related Tools and Resources
More EvvyTools for Analysis
- Graphing Calculator - plot functions and data visually to spot patterns before running tests
- Scientific Calculator - handle the arithmetic behind statistical formulas
- Percentage Calculator - quick percentage computations for proportions and rates
External Resources
- Khan Academy: Statistics and Probability - free, structured curriculum from basics through inference
- NIST/SEMATECH e-Handbook of Statistical Methods - rigorous reference for applied statistics
- Seeing Theory (Brown University) - interactive visualizations that make statistical concepts intuitive
- Wikipedia: Student's t-test - thorough technical reference with derivations and assumptions
Photo by www.kaboompics.com on Pexels
The Math Is the Easy Part
Statistics is not about memorizing formulas. It is about knowing which question to ask and which test answers it. Mean or median? t-test or z-test? Linear or nonlinear? Once you pick the right tool for the question, the calculation itself is mechanical. Paste your data into the Statistics Calculator on EvvyTools, select the analysis, and let the math confirm what your intuition suspects.
