One of the biggest surprises I had while learning data science was realizing that Python isn't the hard part.

You can learn Python in a few weeks. You can become comfortable with pandas pretty quickly. You can even train a machine learning model by following a tutorial.

But none of that means you understand your data.

That's where statistics comes in.

A lot of beginners (myself included) focus on learning tools first because they're exciting. New libraries, dashboards, machine learning models. Statistics often feels like something you can come back to later.

In reality, it's the opposite.

Statistics isn't a side topic in data science. It's the reason the tools work in the first place.

Data only becomes useful when you understand what it represents

Before running any analysis, the first question isn't "Which model should I use?"

It's "What kind of data am I looking at?"

Broadly speaking, data falls into two groups.

Numerical data consists of values you can measure or count. Sales, age, height, temperature.

Categorical data represents labels or groups. Blood type, product category, education level.

That distinction matters more than most beginners realize.

For example, calculating the average blood group doesn't make sense. Treating education levels as though the gap between each level is identical can also lead to misleading conclusions.

Python won't stop you from making those mistakes.

Statistics teaches you when a calculation actually makes sense.

A single number rarely tells the whole story

Imagine two classes that both have an average score of 50.

At first glance, you'd think they performed similarly.

class_A = [48, 49, 50, 51, 52]
class_B = [10, 30, 50, 70, 90]

Both classes have exactly the same mean.

But they're clearly very different.

In Class A, almost everyone performed similarly.

In Class B, performance varied dramatically.

That's why summary statistics come in pairs.

Measures like the mean, median, and mode tell you where the center of the data lies.

Measures like standard deviation, variance, range, and IQR tell you how spread out the data is.

Looking at only one is like reading only half the sentence.

The median isn't a backup plan

When people first learn statistics, they often think:

"Use the mean whenever possible. If it doesn't work, use the median."

That's not really how it works.

The mean uses every value in the dataset, which makes it powerful—but also sensitive to extreme values.

Imagine 99 people earn KES 30,000 each month, while one person earns KES 10 million.

The average income suddenly becomes much higher than what almost everyone actually earns.

The median ignores those extremes and simply finds the middle value.

Sometimes that's a much better description of what's "typical."

Choosing between the mean and median isn't about memorizing rules.

It's about understanding your data.

Outliers aren't always mistakes

One of the first instincts many people have is to delete values that look unusual.

Sometimes that's the right decision.

If someone accidentally entered 250 instead of 25, that's probably a data entry error.

But sometimes the unusual value is exactly what you're looking for.

If you're building a fraud detection system, the suspicious transactions are the most valuable observations in your dataset.

Statistics gives us a systematic way to flag potential outliers using the IQR rule.

Lower fence = Q1 − (1.5 × IQR)

Upper fence = Q3 + (1.5 × IQR)

Anything outside those boundaries is flagged for investigation.

Notice the wording.

Flagged—not automatically deleted.

Statistics helps identify unusual observations.

Context tells you what to do with them.

Machine learning doesn't replace statistics

Every machine learning algorithm makes assumptions.

Linear regression assumes linear relationships and normally distributed residuals.

Naive Bayes assumes features are conditionally independent.

K-Means works best when clusters are reasonably compact and roughly spherical.

If those assumptions don't hold, your model may still produce predictions.

They just won't be reliable.

Understanding statistics helps you know when to trust a model—and when not to.

The part of data science that doesn't become obsolete

Libraries change.

Frameworks change.

The code you write today may look outdated in a few years.

But the important questions stay the same.

• What does this distribution tell me?
• Is this difference meaningful or just random variation?
• Is this outlier important or just an error?
• Can I trust this result?

Those are statistical questions.

And they're the questions that separate someone who can write code from someone who can genuinely analyze data.

If you're starting your journey into data science, don't treat statistics as something to learn later.

It's the foundation that makes everything else make sense.