Univariate Statistics Analyzer

Descriptive Statistics Data Quality

Your first step in any data analysis: understand each variable individually before exploring relationships. Upload marketing data and instantly assess distributions, detect outliers, test for normality, and generate publication-ready summaries.

QUICK START: Choose Your Path

OVERVIEW & WHAT THIS TOOL SOLVES

Jumping straight into regression or A/B testing without inspecting each variable on its own is how a single data-entry outlier, a batch of missing codes, or a heavily right-skewed spend distribution ruins an entire marketing study. The Univariate Statistics Analyzer profiles every column in your dataset one at a time—automatically separating continuous metrics from categorical labels, auditing data quality, testing normality, and letting you stress-test how outliers pull the mean vs. the median.

🎯 Questions It Answers

What is a typical value for this variable (Mean vs. Median vs. Mode), and how spread out are customers around it (SD vs. IQR)?
Is a continuous metric roughly bell-shaped (Normal), or do skewness, heavy tails (kurtosis), and outliers require a log transform or non-parametric test?
How balanced or concentrated are the categories in our segment, channel, or survey-response columns?

📂 What Data You Need

Multi-Column Dataset: Preset marketing case study or uploaded CSV containing continuous and/or categorical columns
Variable Type Override: Automatic detection of Continuous vs. Categorical measurement scales (with 1-click manual override)
Visual & What-If Controls: Switch between Boxplot, Histogram, Violin, Density, and Q-Q plots, or trim outliers live

💡 What You'll Discover

Data Quality Scorecard: Missing value counts, IQR outlier flags, and distribution health badges across all variables
Complete Summary Suite: Mean, SD, Median, IQR, Min/Max, Skewness, Excess Kurtosis, and Shapiro–Wilk normality test
Interactive What-If Sandbox: Live demonstration of how winsorizing or removing extreme points shifts Mean/SD vs. Median/IQR

🔍 Key Nuances & Common Traps to Watch For

Mean & SD vs. Median & IQR on Skewed Spend Data: In e-commerce and B2B revenue, a handful of whale accounts pull the arithmetic mean far above the typical customer. Report the Median and IQR alongside the Mean whenever skewness exceeds |g₁| > 1.
Numeric Codes Are Not Always Continuous: A column containing 1, 2, 3, 4 for Region or Loyalty Tier is a categorical factor—averaging it to "Region 2.4" is meaningless. Use the Variable Type toggle to switch coded categories to Categorical mode.
Shapiro–Wilk Sensitivity in Large Samples: With n > 500, the Shapiro–Wilk test flags tiny, harmless deviations from normality with p < 0.05; always inspect the Q-Q plot and skewness/kurtosis values alongside the p-value.

📐 Under the Hood: Model Equations & Variable Definitions

1. Central Tendency & Dispersion (Sample Mean, Standard Deviation & IQR)

\[ \bar{x} = \frac{1}{n}\sum_{i=1}^{n} x_i, \qquad s = \sqrt{\frac{1}{n - 1}\sum_{i=1}^{n} (x_i - \bar{x})^2}, \qquad \text{IQR} = Q_3 - Q_1 \]

💬 Read it in plain English: The sample mean x̄ is the balance point of all observations and s measures typical distance from that mean, while the interquartile range (IQR) spans the middle 50% of sorted values from the 25th (Q₁) to the 75th (Q₃) percentile.

x_i, n

Observed Values & Valid CountYOUR DATA ⓘ

Individual non-missing numeric observations and valid sample size n.

x̄, s

Sample Mean & Standard DeviationOUTPUT ⓘ

Parametric center and spread (sensitive to extreme tail values).

Q₁, Median, Q₃

Sample Quartiles (25%, 50%, 75%)OUTPUT ⓘ

Order-statistic cutpoints dividing sorted observations into four equal quarters.

IQR

Interquartile RangeOUTPUT ⓘ

Robust spread of the middle 50% of data; points beyond Q₁ − 1.5·IQR or Q₃ + 1.5·IQR are flagged as outliers.

2. Distribution Shape: Sample Skewness (g₁) & Excess Kurtosis (g₂)

\[ g_1 = \frac{1}{n}\sum_{i=1}^{n}\left(\frac{x_i - \bar{x}}{s}\right)^3, \qquad g_2 = \frac{1}{n}\sum_{i=1}^{n}\left(\frac{x_i - \bar{x}}{s}\right)^4 - 3 \]

💬 Read it in plain English: Skewness cubes standardized z-scores to measure asymmetry (positive = right tail, negative = left tail), while excess kurtosis raises z-scores to the fourth power (minus 3 for a normal curve) to measure tail heaviness and outlier proneness.

g₁

Sample SkewnessINFERENCE ⓘ

Asymmetry index (0 = symmetric; > +1 right-skewed; < −1 left-skewed).

g₂

Excess KurtosisINFERENCE ⓘ

Tail-weight index relative to a normal bell curve (0 = normal tails; > +2 heavy/outlier-prone tails).

DATA SOURCE

Load a marketing use case:

📊 Real Marketing Data Scenarios

Select a preset scenario to explore real-world datasets with mixed variable types. Each scenario includes continuous metrics (revenue, engagement scores, visit duration) and categorical dimensions (segments, channels, regions) so you can practice interpreting both types.

Drag & Drop data file (.csv, .tsv, .txt, .xls, .xlsx)

Provide a CSV or TSV with column headers. Up to 5,000 rows per file.

INPUTS & SETTINGS

Variables in Dataset

ℹ️ About Data Types & How to Change Them

We automatically detect whether each variable is Continuous (numeric) or Categorical (labels/groups). Sometimes you may want to override this:

📊 Treat as Continuous when...

Values represent measurements (revenue, age, scores)
Mathematical operations make sense (averages, sums)
You want histograms, box plots, mean/median statistics

🏷️ Treat as Categorical when...

Numbers represent groups or codes (1=Low, 2=Med, 3=High)
There are few unique values that represent categories
You want frequencies, percentages, bar charts

Example: A "Satisfaction" variable with values 1-5 could be treated as continuous (mean = 3.7) or categorical (40% rated it "5"). Click the ✏️ button next to any variable to change its type.

DATA QUALITY DASHBOARD

📋

Total Records —

📊

Variables —

⚠️

Missing Values —

🎯

Potential Outliers —

📋 Variable-by-Variable Data Quality

💡 Data Quality Tips

Missing values: Consider if they're random or systematic. High missing % may indicate a data collection issue.
Outliers: Check if extreme values are real or data entry errors. The IQR method flags values beyond 1.5×IQR from quartiles.
Variable types: Verify the auto-detected types are correct. Use the override buttons if needed.

🎓 Understanding Data "Problems" vs. Data Characteristics

New analysts often panic when they see outliers or non-normal data. But not every statistical "flag" is a problem to fix—some are just characteristics of your data that inform how you analyze it. Learning to tell the difference is a crucial analytical skill.

🟢 EXPECTED PATTERN

Not a problem—just how this type of data behaves

Right-skewed money data: Income, revenue, and purchase amounts are almost always right-skewed. Most customers spend modest amounts; a few big spenders pull the tail right. This is normal for this data type.
Non-normal bounded scales: Rating scales (1-5 stars, 1-10 NPS) can't be perfectly normal because they have floors and ceilings. Pile-ups at extremes are expected.
Legitimate high-value outliers: Your top 1% of customers really do spend 10× more. Enterprise clients really do have 50× more employees. These aren't errors—they're valuable insights.
Bimodal distributions: Seeing two peaks often means you have two distinct customer segments—that's a finding, not a flaw!

What to do: Document the pattern, consider using median instead of mean for central tendency, and choose appropriate statistical tests (non-parametric if needed).

🟡 NEEDS INVESTIGATION

Could be real, could be a problem—dig deeper

Unusually high outlier rate (>5%): If many values are flagged as outliers, either your data has unusual spread or something went wrong during collection.
Unexpected missing patterns: If one variable has 40% missing while others have 2%, investigate why. Was that question optional? Did the form break?
Strange distribution shapes: A sharp cliff at a round number (like many $99.99 values) might indicate data truncation or intentional pricing effects.
Extreme value gaps: If you have values 1-100 and then suddenly a 10,000, investigate that jump before assuming it's valid.

What to do: Look at the raw data. Ask: "Is there a business reason this value exists?" Check with whoever collected the data before making changes.

🔴 LIKELY DATA ERROR

Strong evidence something went wrong—fix before analysis

Impossible negative values: Negative age, negative revenue (unless tracking refunds), negative item counts.
Values outside theoretical bounds: Percentages over 100%, ratings of 6 on a 1-5 scale, ages over 120, probabilities below 0.
Suspicious repetition: Same extreme value repeated many times (copy-paste error?), or values that look like dates in a numeric field (e.g., "20260208").
Unit mismatches: One value in thousands while others are in millions (e.g., $50 vs. $50,000,000 for similar transactions).

What to do: Flag for review. Don't delete automatically—document what you found and how you handled it. Consider if these rows should be excluded or corrected.

STATISTICS & VISUALIZATIONS

Summary Statistics

Visualization type:

📊 How to Read This Chart

🧪 What-If Sandbox

Experiment

What is this? A safe space to experiment with your data and see how statistics change under different conditions.

Why use it? Build intuition about how outliers affect averages, how sample size impacts precision, and whether your findings are robust.

Try these experiments:

Remove outliers — See how much extreme values affect your mean and standard deviation
Change sample size — Watch how statistics "wobble" with fewer data points

🔬 Open Sandbox Controls

📈 Normality Assessment

Many statistical tests assume your data is normally distributed. Use these diagnostics to check.

Shapiro-Wilk Test

Test Statistic (W) —

p-value —

Q-Q Plot (Quantile-Quantile)

💡 How to Interpret Normality Results

Reading the Shapiro-Wilk test:

p ≥ 0.05: No strong evidence against normality. You can reasonably assume the data is normally distributed.
p < 0.05: Evidence that data departs from normality. Consider non-parametric tests or transformations.
W close to 1: Data closely follows a normal distribution.
Note: With large samples (n > 50), even tiny departures from normality become "significant." Use visual inspection alongside the test.

Reading the Q-Q plot:

The dots always rise: they are your values sorted from smallest to largest, so judge them against the straight line, not by their climb.
Dots on the line: close to normal.
Both ends above the line (a smile): right-skewed; a long tail of high values.
Both ends below the line (a frown): left-skewed; a long tail of low values.
Low end below, high end above (an S): heavy tails; more extreme values than expected (leptokurtic).
Low end above, high end below: light tails; fewer extreme values than expected (platykurtic).

💡 How to Interpret These Statistics

Central Tendency (Where's the middle?)

Mean ≈ Median: Data is roughly symmetric—mean is a good summary.
Mean > Median: Right-skewed—outliers on high end pull mean up. Median may be more representative.
Mean < Median: Left-skewed—outliers on low end pull mean down.

Spread (How variable is the data?)

Coefficient of Variation (CV = SD/Mean): CV < 0.5 is low variability, > 1 is high variability.
IQR vs Range: If Range >> 4×IQR, you likely have extreme outliers.
SD interpretation: In a normal distribution, ~68% of values fall within ±1 SD of mean, ~95% within ±2 SD.

Shape (Distribution patterns)

|Skewness| < 0.5: Approximately symmetric.
|Skewness| 0.5–1: Moderately skewed.
|Skewness| > 1: Highly skewed—consider transformations or non-parametric methods.
|Kurtosis| > 2: Heavy tails with more outliers than normal distribution.

Univariate Statistics Analyzer

👨‍🏫 Professor Mode: Guided Learning Experience

QUICK START: Choose Your Path

Try an Example

Upload My Data

Learn the Basics

OVERVIEW & WHAT THIS TOOL SOLVES

🎯 Questions It Answers

📂 What Data You Need

💡 What You'll Discover

1. Central Tendency & Dispersion (Sample Mean, Standard Deviation & IQR)

2. Distribution Shape: Sample Skewness (g₁) & Excess Kurtosis (g₂)

DATA SOURCE

Use a Case Study

📊 Real Marketing Data Scenarios

Upload Your Data

INPUTS & SETTINGS

Variables in Dataset

📊 Treat as Continuous when...

🏷️ Treat as Categorical when...

DATA QUALITY DASHBOARD

🔎 Issues Detected in Your Data

STATISTICS & VISUALIZATIONS

Summary Statistics

🧪 What-If Sandbox

📈 Normality Assessment

Shapiro-Wilk Test

Q-Q Plot (Quantile-Quantile)

Summary Tables

APA-Style Statistical Reporting

Managerial Interpretation

🔮 WHAT'S NEXT? — From Univariate to Deeper Analysis

🏆 CHALLENGE MODE — Test Your Understanding

Analysis Checklist

My Findings

🏆 Challenge Mode