In this first lesson you will discover what statistics really is and why it has become an essential skill in the professional world. Far from being a collection of abstract formulas, statistics is the set of methods that lets us turn data into decisions. You will learn the difference between describing data and inferring conclusions, understand what a population and a sample are, and distinguish between parameters and statistics. You will also meet NovaMarket, the (fictional) company where you will work as a data analyst throughout the course: every example and exercise revolves around its data.

1. Welcome to NovaMarket: your job for the duration of this course
2. What is statistics, and what is it for in business?
3. Descriptive and inferential statistics
4. Population and sample
5. Parameter and statistic
6. Units of analysis and observations

Imagine that today is your first day as a data analyst at the head office of NovaMarket, a Spanish supermarket chain with:

• 42 physical stores spread across the country.
• An online store with home delivery.
• A loyalty program called "Club Nova", with hundreds of thousands of registered customers.
• Regular promotional campaigns (flyers, discounts, personalized coupons).

Your direct manager is Marta, head of analytics. Marta receives questions from across the organization and turns them into assignments for your team. Here are some real examples of the kind of requests that arrive every week:

Notice something important: none of these questions mentions the word "statistics", yet every one of them needs statistics to be answered rigorously. That is exactly your job: translating business questions into data questions, and data answers into business decisions.

Throughout the course, Marta will keep handing you assignments of increasing difficulty. In this module we start with the essentials: understanding the language of statistics.

Statistics is the science of collecting, organizing, summarizing, analyzing and interpreting data in order to draw conclusions and support decision-making under uncertainty.

Let's unpack that definition, because each verb corresponds to a phase of an analyst's real work:

1. Collecting: obtaining the data (purchase receipts, surveys, delivery records...). We cover this in the lesson Data Collection and Sampling.
2. Organizing: structuring the data into clean, consistent tables.
3. Summarizing: reducing thousands of numbers to a handful of understandable indicators (means, percentages, charts). This is the subject of Module 2.
4. Analyzing: looking for patterns, relationships and differences.
5. Interpreting: answering the original business question, and stating how much confidence the answer deserves.

The last part of the definition — under uncertainty — is the key that separates statistics from mere bookkeeping. At NovaMarket you will almost never have all the data you would like:

• You cannot survey all 380,000 Club Nova members; you will survey a few thousand.
• You cannot know how much the Zaragoza store will sell next month; you can only estimate it.
• You cannot observe every potential customer in Spain; only the ones who already shop with you.

Statistics gives you the tools to say sensible things about what you cannot see based on what you can see, and to quantify the risk of being wrong.

Even if you never compute anything by hand in your job, understanding statistics allows you to:

• Ask better questions: "Is that 3% increase meaningful, or is it within normal variation?"
• Spot hasty conclusions: a report comparing two stores using data from a single week probably proves nothing.
• Communicate rigorously: distinguishing between "sales went up" and "sales went up in the sample we analyzed, with such-and-such margin of error".
• Talk to technical colleagues: understanding what a data scientist or a business intelligence tool is telling you — and what it is not.

Statistics is traditionally divided into two major branches. Understanding the difference is fundamental because it determines what you can legitimately claim with your data.

Descriptive statistics organizes and summarizes the data you already have, without trying to go beyond them. It answers questions of the form "what do these data look like?".

An example at NovaMarket: Marta asks you for a report on yesterday's sales across the 42 stores. You have the exact figure for every store, so you can compute:

• Total sales for the day: €1,847,300.
• Average sales per store: \( 1{,}847{,}300 / 42 \approx 43{,}983 \) euros.
• The top-selling store (Madrid-Centro, €91,200) and the lowest (Cuenca, €18,400).

There is no uncertainty here: you are describing exactly what happened. You are not estimating anything; you are summarizing.

Inferential statistics uses data from a part (a sample) to draw conclusions about the whole (the population), attaching to each conclusion a measure of its reliability. It answers questions of the form "what can I conclude about what I have not observed?".

An example at NovaMarket: Marta wants to know the satisfaction of all Club Nova members, but you have only surveyed 1,200 of them. If the mean satisfaction in the survey is 7.8 out of 10, statistical inference lets you make a claim like: "the mean satisfaction of all members lies, with high confidence, between 7.6 and 8.0". That leap — from 1,200 respondents to 380,000 members — is an inference, and doing it properly is the subject of Modules 3 to 5.

A very common mistake in business is to present descriptive results from a sample as if they were conclusions about the whole population, with no measure of uncertainty whatsoever. This course will teach you not to fall into that trap.

These two concepts are the backbone of all inferential statistics.

• Population: the complete set of elements we want to draw conclusions about. It does not have to be people: it can be transactions, deliveries, products or stores.
• Sample: the subset of the population that we actually observe or measure.

\[ \text{Sample} \subset \text{Population} \]

The population is defined by the business question, not by whatever data you happen to have. Let's see this with real assignments from Marta:

Look closely at the last case: when you can observe the entire population (there are only 42 stores and you have the figure for all of them), there is nothing to infer; describing is enough. That is called working with a census. In the lesson Data Collection and Sampling we will see when a census pays off, when a sample does, and how to choose that sample well.

Because observing the complete population is usually:

• Expensive: surveying 380,000 members would cost a fortune.
• Slow: by the time you finished, the results would be out of date.
• Impossible: the population may include future elements (the rest of the year's deliveries do not exist yet).
• Unnecessary: a well-chosen sample of a few thousand can give results almost as precise as a census.

The trade-off is that a sample introduces uncertainty: a different sample would have produced slightly different numbers. Managing that uncertainty is precisely the job of inferential statistics.

When we compute a summary value (a mean, a percentage...), what we call it depends on what it is computed over:

• Parameter: a value computed over the complete population. It is usually unknown (that is why we sample) and is written with Greek letters: \( \mu \) (population mean), \( \sigma \) (population standard deviation), \( \pi \) or \( p \) (population proportion).
• Statistic: a value computed over the sample. It is known (you compute it from your data) and is written with Latin letters: \( \bar{x} \) (sample mean), \( s \) (sample standard deviation), \( \hat{p} \) (sample proportion).

The relationship between the two is the heart of inference:

\[ \underbrace{\bar{x}}{\text{statistic (known)}} ; \longrightarrow ; \text{estimates} ; \longrightarrow ; \underbrace{\mu}{\text{parameter (unknown)}} \]

Marta wants to know the mean monthly spend of all Club Nova members (that is the parameter \( \mu \), unknown). Since analyzing all of them today is not feasible, you draw a random sample of 8 members and look up their spend over the last month:

Step 1. Add up the spend figures:

\[ 210 + 145 + 320 + 95 + 260 + 180 + 155 + 235 = 1{,}600 \text{ euros} \]

Step 2. Divide by the number of members in the sample (\( n = 8 \)):

\[ \bar{x} = \frac{1{,}600}{8} = 200 \text{ euros} \]

Interpretation. The statistic \( \bar{x} = \) €200 is your best estimate of the parameter \( \mu \). But be careful: had you picked a different 8 members, you would have obtained a different mean (perhaps €187, perhaps €214). The true value of \( \mu \) remains unknown; what you will learn in Module 5 is how to build statements of the form "\( \mu \) most likely lies between this value and that one". For now, hold on to the distinction:

A classic mnemonic: P with P and S with S — the Parameter describes the Population; the Statistic summarizes the Sample.

Before computing anything, an analyst must be absolutely clear about what each row of the data represents. Two concepts come into play:

• Unit of analysis (or statistical unit): the type of individual element you are studying — a customer, a transaction, a store, a delivery.
• Observation: the data recorded for one specific unit; in a table, each observation is one row.

On each unit, one or more characteristics are measured (the receipt amount, the store's province...). Those characteristics are the variables, and they occupy the columns of the table; we will study them in detail in the next lesson, Types of Data.

The same question changes its answer depending on the unit of analysis you choose. Suppose Marta asks: "what is our average amount?".

• If the unit is the transaction: you add up all receipt amounts and divide by the number of transactions. You get the average transaction value (say, €32.40).
• If the unit is the customer: you add up each customer's total spend for the month and divide by the number of customers. You get the average spend per customer (say, €187).
• If the unit is the store: you get the average sales per store (say, €43,983 per day).

All three numbers are correct, but they answer different questions. Part of the analyst's craft is asking Marta (or whoever placed the request): "average per what: per transaction, per customer or per store?".

Here is what an extract from the e-commerce deliveries table might look like, where the unit of analysis is the delivery:

• Each row is an observation (one specific delivery).
• Each column is a variable measured on that delivery.
• If this table contains 500 of the year's 120,000 deliveries, it is a sample; the 120,000 are the population.

• Confusing describing with inferring. "62% of respondents prefer our own brand" is not the same as "62% of our customers prefer our own brand". The first is a description of the sample; the second is an inference that requires methods (and caution) that we will cover later on.
• Defining the population poorly. If you only survey Club Nova members, your population is "the members", not "all NovaMarket customers". Customers without a loyalty card may be very different.
• Believing that more data always means population. Having millions of receipts does not mean you have the population: if your question is about customers' future behavior, even every past transaction is, conceptually, a sample.
• Mixing units of analysis. Comparing one store's "average spend per transaction" with another store's "average spend per customer" is comparing apples and oranges. Always fix the unit before computing.
• Professional tip: when you receive an assignment, write down in one sentence the population, the unit of analysis and what you want to learn about it. Thirty seconds of definition save hours of misdirected analysis.

Marta forwards you this message from the head of logistics: "We analyzed 400 July deliveries chosen at random and the mean time was 28.3 hours. I want to know how this year's delivery service is doing."

1. What is the population of the study?
2. What is the sample?
3. Is the value 28.3 hours a parameter or a statistic?
4. Is the claim "the mean time of the 400 deliveries we analyzed was 28.3 hours" descriptive or inferential? What about "our service delivers in about 28 hours on average"?

For each assignment, state the most appropriate unit of analysis:

1. "What percentage of transactions includes at least one own-brand product?"
2. "What is the average tenure of Club Nova members?"
3. "How many stores exceeded one million euros in sales in August?"
4. "What proportion of online deliveries arrives in under 24 hours?"

In a meeting, a marketing manager states: "We asked 50 customers outside the Madrid-Centro store and 80% of them like the new flyer, so 80% of NovaMarket's customers like the new flyer." Point out two statistical problems with this claim, using the concepts from this lesson.

Solution 1

1. Population: all of NovaMarket's e-commerce deliveries this year (including those that have not happened yet, if the question is about "this year's service"; at a minimum, all July deliveries if we restrict ourselves to the month studied).
2. Sample: the 400 July deliveries analyzed.
3. It is a statistic (\( \bar{x} = 28.3 \) h): it was computed over the sample. The parameter would be the mean time of all deliveries, which we do not know.
4. The first claim is descriptive: it speaks only of the 400 observed deliveries and is an exact fact. The second is inferential: it generalizes to the whole service. Common mistake: treating them as equivalent; the second requires justifying that the sample is representative and attaching a margin of error to the 28, which we will learn to do in Module 5.

Solution 2

1. The transaction (each row is a receipt; you check whether it contains own-brand product).
2. The Club Nova member.
3. The store.
4. The delivery.

Common mistake: in question 1 many people answer "the product" or "the customer". Look at the wording: it asks for a percentage of transactions, so the element being counted (and forming the rows) is the transaction.

Solution 3

Two main problems (any one of them, well argued, makes the answer valid; ideally you spot both):

1. A leap from description to inference without safeguards: the 80% is a statistic of the sample; presenting it as the value for "NovaMarket's customers" treats a statistic as if it were the parameter, with no margin of error at all (and with only 50 people, that margin would be large).
2. A poorly represented population: the survey took place at a single store (Madrid-Centro) and only among people leaving after shopping. That sample does not represent the customers of the 42 stores, nor those of the online channel, so generalizing to "NovaMarket's customers" is unjustified. (We will see what this problem is called and how to avoid it — selection bias — in lesson 01-03.)

Tip: you do not yet need to know how to compute margins of error to catch these flaws; it is enough to always ask yourself "what population am I talking about, and who did I actually observe?".

In this lesson you have laid the foundations for the whole course. You now know that statistics exists to support data-driven decisions under uncertainty, and you have mastered its essential vocabulary: descriptive statistics summarizes what was observed and inferential statistics generalizes to what was not; the population is the whole you want to draw conclusions about and the sample is the part you observe; the parameter is the population value (unknown, in Greek letters) and the statistic is its sample estimate (known, in Latin letters); and every data table is organized into observations (rows) of a well-defined unit of analysis. You have also met NovaMarket and Marta, who will keep bringing you assignments in every lesson. The next step is to examine the columns of those tables: in the next lesson, Types of Data, you will learn to classify variables and to know which operations each one allows — the foundation for never, ever computing the "mean" of a postal code.