Statistics Basics for Data Analysis: Averages, Variance, Correlation and Significance

Your dashboard says average order value went up 6% last month. Your marketing lead says the new email “drove” more sales. A vendor shows you a test result that is “statistically significant.” Should you believe any of it?

You don’t need a statistics degree to answer that. Most everyday business analysis rests on 4 ideas: what is typical, how spread out the data is, whether 2 things move together, and whether a difference is real or just luck. This guide explains each one with plain business examples and the spreadsheet functions that calculate them.

If you are building toward an analyst role, this is the foundation that sits under every tool. Pair it with our guides on what data analytics is and how to become a data analyst.

Why statistics matters for business data

Raw numbers mislead more often than they inform. A single big order can inflate an average. Two charts can look connected when they are not. A test can “win” because of random variation in who happened to visit your site that week.

Statistics gives you a short list of checks before you act on a number:

• Is this figure typical, or pulled by a few extreme values?
• How much does this number normally bounce around?
• Do these 2 measures actually move together, and how strongly?
• Is this difference bigger than what chance alone would produce?

Those 4 questions map neatly to the 4 sections below.

Averages: mean, median and mode

The mean is what most people call the average: add up the values and divide by how many there are. The median is the middle value once you sort the data, so half the values sit below it and half above. The mode is the value that appears most often. The NIST/SEMATECH e-Handbook of Statistical Methods lists these 3 as the common ways to describe the “center” of a dataset.

The mean and median often tell different stories. Take a small online store with 10 orders in a day: 9 orders of about $40 and 1 wholesale order of $1,200. The mean order value comes out near $156. The median is $40. If you plan inventory or marketing around “the average customer spends $156,” you will be badly wrong.

When to use which

Measure	What it tells you	Best for	Watch out for
Mean	Arithmetic average	Totals and forecasts, data without big outliers	Pulled up or down by extreme values
Median	Middle value	Skewed data such as order values, salaries, house prices, time on site	Hides what is happening at the extremes
Mode	Most common value	Categories such as most popular size or plan	Can be unstable in small samples

A practical rule: when data is skewed (a long tail of very large or very small values), report the median alongside the mean. Revenue per customer, deal size and support ticket resolution time are almost always skewed.

In Excel or Google Sheets, use AVERAGE, MEDIAN and MODE.SNGL. For a quick summary of a whole column, a pivot table can show mean, count, min and max side by side.

Spread: range, variance and standard deviation

Two stores can both average 100 orders a day. One sells 95 to 105 orders every day. The other swings between 20 and 180. Same average, very different businesses to staff and stock. Spread captures that difference.

• Range is the maximum minus the minimum. It is easy but driven entirely by the 2 most extreme values.
• Variance measures the average squared distance of each value from the mean. Squaring makes every distance positive and gives big deviations more weight.
• Standard deviation is the square root of variance. It is the one you will actually use, because it is in the same units as your data (dollars, orders, minutes).

If daily orders have a mean of 100 and a standard deviation of 8, most days land fairly close to 100. A standard deviation of 40 means planning around “100 a day” is risky.

Sample vs population

Excel has 2 standard deviation functions. Microsoft’s documentation explains that STDEV.S estimates standard deviation from a sample and uses the “n-1” method, while STDEV.P is for when your data is the entire population. In business, you are usually working with a sample (last quarter’s customers stand in for customers in general), so STDEV.S is the safer default.

Using spread in practice

Standard deviation helps you spot unusual days. A common rule of thumb flags values more than 2 or 3 standard deviations from the mean for a closer look. That is how many simple anomaly alerts on a business dashboard work. It is not proof something is wrong, just a signal to check for a data error, a promotion or a real problem.

Key takeaway: Never report an average on its own. Pair it with the median when data is skewed and with the standard deviation so readers know how much the number normally moves.

Correlation: do 2 things move together

Correlation measures how closely 2 numeric variables move together. The most common version, Pearson’s correlation coefficient, ranges from -1 to +1. Microsoft’s CORREL function returns it in Excel, and values near +1 or -1 indicate a strong relationship while values near 0 indicate little linear relationship.

Correlation value	Rough reading	Business example
+0.7 to +1.0	Strong positive	Ad spend and website sessions often rise together
+0.3 to +0.7	Moderate positive	Temperature and iced drink sales at a cafe
-0.3 to +0.3	Weak or none	Customer zip code digits and order value
-0.7 to -1.0	Strong negative	Delivery time and repeat purchase rate, in some businesses

These bands are rules of thumb, not fixed laws. What counts as “strong” depends on your field and how noisy the data is.

Correlation is not causation

This is the most important caveat in business analytics. Ice cream sales and sunburn cases rise together, but ice cream does not cause sunburn. Hot weather drives both. That hidden driver is called a confounding variable.

In your own data, watch for:

• Seasonality. Many metrics rise in Q4 together simply because it is holiday season.
• Growth. If your business is growing, almost every volume metric correlates with every other.
• Reverse cause. Loyal customers may open more emails because they are loyal, not the other way round.

To show cause and effect, you need a controlled experiment, such as an A/B test, or careful methods from predictive analytics and econometrics that account for confounders.

Correlation also only captures straight-line relationships. Always plot the data in a scatter chart before trusting a single coefficient.

Statistical significance and p-values

When you run an A/B test, the 2 versions will almost never perform identically. The question is whether the gap is bigger than you would expect from random variation alone. That is what statistical significance tries to answer.

A p-value is the probability of seeing a difference at least as large as the one you observed if there were truly no difference between the versions. A small p-value means your result would be surprising under the “no difference” assumption. Many teams use 0.05 as a cutoff, but that threshold is a convention, not a law of nature.

In Excel, T.TEST returns the p-value for a Student’s t-test. You choose 1 or 2 tails and the test type: paired, 2-sample equal variance or 2-sample unequal variance.

What the experts say p-values do not mean

In March 2016, the American Statistical Association published its first-ever formal statement on statistical practice, setting out 6 principles about p-values. In plain terms:

1. A p-value indicates how incompatible the data are with a specified statistical model.
2. It is not the probability that your hypothesis is true or that the result happened by chance alone.
3. Business or policy decisions should not rest only on whether a p-value passes a threshold.
4. Proper inference needs full reporting and transparency.
5. A p-value does not measure the size or importance of an effect.
6. On its own, a p-value is not a good measure of evidence.

For business readers, point 5 matters most. A test with a huge sample can find a “significant” lift of 0.1% that is not worth the engineering time. Always ask for the effect size (how big the difference is) and a confidence interval (the plausible range for that difference), not just “significant or not.”

Common testing mistakes

• Peeking. Checking results daily and stopping the moment they look significant inflates false wins. Decide your sample size or test duration up front.
• Testing too many things. Run 20 comparisons at a 0.05 cutoff and you should expect about 1 false positive by chance.
• Tiny samples. A handful of conversions can swing wildly. Wait for enough data.
• Ignoring segments after the fact. Slicing results until something “wins” is a recipe for false discoveries.

Sampling and bias: the quiet problem

Every statistic above assumes your data represents what you care about. Often it does not.

• Survivorship bias. Analyzing only current customers ignores the ones who left, which hides the reasons they left.
• Selection bias. Survey responses come from people who chose to answer, who may be happier or angrier than average.
• Small samples. Results from 30 customers can differ a lot from results from 3,000.
• Data quality. Duplicates, missing values and inconsistent formats distort every calculation. Run through a data cleaning process before you analyze.

Before trusting any number, ask: who or what is missing from this data?

Where to start: a 6-step routine for any dataset

Use this routine whenever you get a new spreadsheet or report:

1. Count and clean. Check the row count, remove duplicates and note missing values.
2. Describe the center. Calculate mean and median. If they differ a lot, the data is skewed, so lead with the median.
3. Describe the spread. Calculate standard deviation, min and max. Flag values beyond 2 to 3 standard deviations for review.
4. Plot it. A histogram shows the shape of 1 variable. A scatter chart shows the relationship between 2.
5. Check relationships. Use CORREL for pairs you care about, then ask what confounders could explain the pattern.
6. Test before you claim. For before-and-after or A/B comparisons, check significance, effect size and sample size together.

Quick reference: spreadsheet functions

Task	Excel / Google Sheets function
Mean	AVERAGE(range)
Median	MEDIAN(range)
Most common value	MODE.SNGL(range)
Sample standard deviation	STDEV.S(range)
Sample variance	VAR.S(range)
Correlation	CORREL(range1, range2)
t-test p-value	T.TEST(range1, range2, tails, type)

If you want to move beyond spreadsheets, our SQL for beginners guide shows how to calculate averages and counts directly in a database, and the glossary defines terms you will meet along the way.

Conclusion

Statistics for data analysis comes down to 4 habits: check what is typical, check how much it varies, test whether relationships are real, and never treat a p-value as the whole story. Those habits catch most of the mistakes that lead to bad decisions, from planning around a skewed average to shipping a “winning” test that was just noise. Start with the 6-step routine on your next report, and the numbers will start telling you something you can trust.

Sources

• NIST/SEMATECH e-Handbook of Statistical Methods: Measures of Location — definitions of mean, median and mode and how skewed data affects them
• American Statistical Association: Statement on Statistical Significance and P-Values (2016) — the 6 principles on interpreting p-values
• Microsoft Support: STDEV.S function — sample vs population standard deviation in Excel
• Microsoft Support: CORREL function — calculating the correlation coefficient
• Microsoft Support: T.TEST function — tails and test types for t-test p-values