📘 SERIES · CHAPTER 69📐 STATISTICS

Statistics Fundamentals for Data Analysts — Mean, Distributions, Correlation & Hypothesis Testing

You do not need a statistics degree to be a great data analyst. You do need to understand the foundational concepts well enough to avoid misleading your stakeholders. This chapter covers the essential statistics every analyst uses in their day-to-day work — no heavy calculus, just practical concepts with real examples.

⏱ 22 min read📅 September 2026✍ EVIKA ACADEMY, Noida

Why Every Data Analyst Needs Statistics

Statistics is the language your data speaks. Without it, you can describe what happened in your data, but you cannot say whether it was significant, whether it will repeat, or whether two things are truly related. The most common analyst errors — confusing correlation with causation, averaging rates incorrectly, drawing conclusions from tiny samples — are all statistics errors, not technical errors.

📊
Descriptive stats
Summarise what happened in your data — mean, median, percentiles, standard deviation.
🎲
Probability
Understand how likely events are and how distributions describe real-world patterns.
🔗
Correlation & causation
Know when two variables move together vs when one actually causes the other.
🧪
Hypothesis testing
Decide whether differences you observe are real or just random noise.

Descriptive Statistics — Summarising Your Data

Measures of Central Tendency

Mean (Average)
Sum ÷ Count
Use when: Works well when data is symmetrically distributed without extreme outliers.
Caution: A single extreme value can drag the mean far from where most data sits. Five salaries of ₹5L and one of ₹5Cr gives a mean of ₹88L — not representative.
Median
Middle value when sorted
Use when: Use for skewed data: salaries, house prices, wait times, income distributions.
Caution: The median ignores the magnitude of extreme values entirely — useful for robustness, but misses information about the tails.
Mode
Most frequent value
Use when: Useful for categorical data: most popular product, most common city, most frequent support issue.
Caution: Meaningless for continuous numerical data where almost every value is unique.

Measures of Spread

RangeMax − Min

Gives the total spread but is extremely sensitive to a single outlier. Useful as a quick sanity check: if range is unexpectedly large, investigate data quality.

💡 Order values: ₹50 to ₹1,20,000 — range tells you there are very different customer segments.

VarianceAverage of squared deviations from the mean

Used internally in statistical formulas. Hard to interpret directly because the unit is squared (e.g., ₹²). Always reported alongside standard deviation.

💡 Rarely quoted to stakeholders directly.

Standard Deviation (SD)√Variance — spread in original units

The most used spread measure. About 68% of data falls within 1 SD of the mean, 95% within 2 SD (for normal distributions). High SD = widely spread data; low SD = tightly clustered.

💡 Delivery times: mean 3 days, SD 0.5 days → most deliveries between 2.5 and 3.5 days. SD of 2 days → very inconsistent.

Percentiles / IQRP25, P50, P75 — IQR = P75 − P25

Robust to outliers. The interquartile range covers the middle 50% of data. P90 / P95 / P99 are essential for performance metrics: "99% of page loads under 2s."

💡 Customer support: median resolution 4h, P95 resolution 24h. The P95 is what SLA agreements are built on.

Probability Distributions Data Analysts Actually Use

A distribution describes the shape of your data — how values are spread across the range. Understanding distributions helps you choose the right statistical tests and spot anomalies.

🔔 Normal Distribution (Bell Curve)
When you see it: Heights, weights, exam scores, measurement errors
68-95-99.7 rule: 68% within ±1SD, 95% within ±2SD, 99.7% within ±3SD. Mean = Median = Mode.
Analyst use: Many statistical tests assume normality. Always check if your data is normally distributed before applying tests like t-tests.
📈 Right-Skewed Distribution
When you see it: Salaries, house prices, sales revenue, website session duration
Long tail on the right. Mean > Median. A few very high values pull the mean up.
Analyst use: Use median and IQR instead of mean and SD. Log-transform before modelling. Common mistake: reporting mean salary when median is much lower.
▬ Uniform Distribution
When you see it: Random number generation, dice rolls, random sampling
Every value equally likely. Flat shape. Mean sits exactly in the middle.
Analyst use: Baseline for testing randomness. If user IDs are uniform, random assignment to A/B groups is valid.
✅ Binomial Distribution
When you see it: Click / no-click, convert / not convert, pass / fail
Models the number of successes in N independent trials, each with probability p.
Analyst use: Core of A/B testing and conversion rate analysis. "If baseline conversion is 10%, what is the probability of getting 120 conversions from 1000 users by chance?"
⏱ Poisson Distribution
When you see it: Arrivals per hour, defects per batch, support tickets per day
Models the number of events in a fixed time/space interval when events happen independently at a constant rate.
Analyst use: "Average 5 support tickets per hour. What is the probability of getting 10+ in the next hour?" Essential for capacity planning.

Correlation vs Causation — The Most Important Distinction in Analytics

Correlation means two variables move together. Causation means one variable directly causes the other to change. Confusing these two is one of the most damaging errors an analyst can make — it leads to decisions that do nothing, or worse, create harm.

❌ Classic Correlation Traps
Ice cream sales and drowning rates both spike in summer → ice cream does not cause drowning. Both are caused by a third variable (hot weather = confounding variable).
Countries with more TVs per household have higher life expectancy → TVs do not cause longer life. Both correlate with wealth.
Users who log in more frequently have higher retention → you cannot fix retention just by forcing more logins. High-quality users both log in more AND retain more.

Correlation Coefficient (r) — Quick Reference

r value rangeStrengthDirectionExample
0.9 to 1.0Very strongPositiveAd spend vs revenue (same channel)
0.7 to 0.9StrongPositiveStudy hours vs exam score
0.4 to 0.7ModeratePositiveCustomer age vs average order value
0.0 to 0.4WeakPositiveShoe size vs salary
-0.4 to 0.0WeakNegativeTemperature vs hot beverage sales
-0.7 to -0.4ModerateNegativePrice increase vs units sold
-1.0 to -0.7StrongNegativeSupport wait time vs satisfaction score
Remember: r tells you the strength and direction of a linear relationship only. Two variables can have a strong non-linear relationship with r ≈ 0. Always plot the scatter before interpreting r.

Hypothesis Testing — Making Decisions with Confidence

Hypothesis testing is the formal process for deciding whether an observed difference in data is real or just random chance. Every A/B test result, every "our new feature increased conversions by 12%" claim, should go through this framework.

The 5-Step Hypothesis Testing Process

1
State the null hypothesis (H₀)
H₀ is the assumption that nothing is happening — there is no difference, no effect. Example: "The new button colour has no effect on click rate." You are trying to find evidence AGAINST H₀.
2
State the alternative hypothesis (H₁)
H₁ is what you believe might be true. Example: "The new button colour increases click rate." This is what you are trying to prove.
3
Choose significance level (α)
α is the threshold for deciding results are "significant." The standard is α = 0.05 (5%). This means: if H₀ is true, you would only see results this extreme 5% of the time by chance.
4
Calculate the p-value
p-value = the probability of observing your results (or more extreme) if H₀ is true. A low p-value means your results are unlikely under H₀. If p < α (e.g., p = 0.02 < 0.05), you reject H₀.
5
Interpret the result
p < α → reject H₀ → the difference is statistically significant. p ≥ α → fail to reject H₀ → not enough evidence. NOTE: "statistically significant" does not mean practically significant. A 0.01% improvement can be statistically significant with a million-row dataset.

Common Hypothesis Tests and When to Use Them

TestUse whenAnalyst example
t-test (one-sample)Comparing a sample mean to a known valueIs average order value significantly different from the ₹1,500 target?
t-test (two-sample)Comparing means of two independent groupsDid Group A or Group B spend more per session in an A/B test?
Chi-square testComparing proportions / categorical countsIs conversion rate different between mobile vs desktop users?
ANOVAComparing means across 3+ groups at onceDo conversion rates differ across 4 marketing channels?
Correlation testTesting if r is significantly different from 0Is the relationship between ad spend and revenue statistically significant?

A/B Testing — Applying Hypothesis Testing in Practice

A/B testing is the most common application of hypothesis testing in business analytics. You show version A to one group and version B to another, then test whether the difference in outcome is statistically significant.

A/B TEST EXAMPLE — NEW CHECKOUT BUTTON
CONTROL (Version A)
Grey "Submit" button
Visitors: 5,000
Conversions: 420
8.4%
VARIANT (Version B)
Green "Buy Now" button
Visitors: 5,000
Conversions: 510
10.2%
H₀: The two buttons have equal conversion rates
H₁: Version B has a higher conversion rate
Chi-square test → p-value = 0.003
p = 0.003 < 0.05 → Reject H₀ → Version B is significantly better. Ship the green button.

A/B Testing Pitfalls Analysts Must Avoid

⚠️ Peeking at results too early
Checking significance every day and stopping the moment p < 0.05 inflates false positives. Decide your sample size in advance and run the test to completion.
⚠️ Running too many variants at once
Testing A vs B vs C vs D vs E means some will appear significant purely by chance (multiple comparisons problem). Test one variant at a time or use Bonferroni correction.
⚠️ Ignoring practical significance
A 0.2% lift in conversion is statistically significant on 10 million users but generates ₹800/day in extra revenue. Not worth shipping. Always report absolute lift alongside p-value.
⚠️ Not ensuring random assignment
If users who visit on weekends see Version B and weekday visitors see Version A, you are not testing the button — you are testing weekend vs weekday behaviour.

Statistics in Your Analyst Tools — Quick Reference

Excel
=AVERAGE(), =MEDIAN(), =MODE()
=STDEV() for standard deviation
=PERCENTILE() for P25/P75/P90
=CORREL() for correlation coefficient
Data Analysis ToolPak → t-test, ANOVA
=NORM.DIST() for normal distribution
Python (pandas / scipy)
df.describe() — all descriptive stats
df[col].std(), .var()
df[col].quantile([.25,.5,.75])
df.corr() — full correlation matrix
scipy.stats.ttest_ind(a, b)
scipy.stats.chi2_contingency(table)
Power BI (DAX)
AVERAGE(), MEDIAN()
STDEV.P(), STDEV.S()
PERCENTILE.INC()
CORR() via R/Python visual
Quick Insights → built-in statistical analysis
Q&A → "show distribution of Sales"

Learn Applied Statistics at EVIKA ACADEMY

Our data analytics course in Noida Sector 51 teaches statistics as it is actually used in the workplace — in Excel, Python, and Power BI, with real business datasets. Online & offline classes available.

📱 Book Free Demo Class →

Frequently Asked Questions

What statistics do data analysts actually need to know?

Data analysts need descriptive statistics (mean, median, standard deviation, percentiles), an understanding of common distributions (normal, skewed, binomial), correlation and causation concepts, and the basics of hypothesis testing including p-values and significance levels. Deep knowledge of advanced statistical modelling is a data scientist skill — analysts focus on application over theory.

What is the difference between mean and median in data analysis?

Mean is the arithmetic average and works well for symmetrically distributed data with no extreme outliers. Median is the middle value and is robust to outliers. Use mean for symmetric data like exam scores. Use median for skewed data like salaries, house prices, or income — where a few very high values would make the mean unrepresentative of the typical value.

What does a p-value mean in simple terms?

The p-value is the probability that you would see results this extreme (or more extreme) if there was actually no real effect. A low p-value (typically below 0.05) means it is unlikely that the observed difference is due to random chance, so you conclude the effect is statistically significant. A high p-value means you do not have enough evidence to conclude there is a real difference.

What is the difference between correlation and causation?

Correlation means two variables tend to move together — when one goes up, the other goes up (or down). Causation means one variable directly causes the other to change. The most important rule in data analysis: correlation does not imply causation. A third confounding variable often causes both to move together. Always ask "what else could explain this relationship?" before concluding that A causes B.

Where can I learn statistics for data analytics in Noida?

EVIKA ACADEMY in Noida Sector 51 teaches applied statistics as part of its data analytics courses. Topics include descriptive statistics, distributions, correlation, hypothesis testing and A/B test interpretation — all applied in Excel, Python and Power BI with real business datasets. Online and offline classes are available. WhatsApp +91-8081035456 to book a free demo.

🎓 Free Demo Class — Online & Offline · Noida Sector 51