Chapter 1: Data Analysis

Welcome to Last Minute Lecture.

This free chapter overview is designed to help students review and understand key concepts.

These summaries supplement, not replace, the original textbook and may not be redistributed or resold.

For complete coverage, always consult the official text.

Welcome to the Deep Dive.

We're your shortcut to getting genuinely well -informed, breaking down the key insights from important sources.

And today we're diving into something that, while it's everywhere, data.

It really is.

Poll results, medical studies, stocks, test scores.

It just floods our lives daily.

But it's not just noise, right?

This data is trying to tell us something, a story.

Exactly.

And statistics.

That's the language we need to understand that story.

Our mission here is pretty simple.

Help you decode that story, give you those aha moments, maybe some surprising facts without drowning you in information.

And today we're doing a really focused deep dive, chapter one of the practice of statistics, the sixth edition by Starnes and Taber.

Ah, yes, a classic.

And this is specifically geared towards you AP statistics learners out there.

Our goal, demystify data analysis, one concept at a time.

So let's jump right in.

What exactly is statistics when you boil it down?

Well, at its heart, statistics is both a science and, you know, an art.

Science and art.

Okay.

Yeah, the science of collecting and analyzing data and the art of drawing meaningful conclusions from it.

It's not just number crunching.

It's about using those numbers to make smarter decisions, understand the world better.

Precisely.

Spotting patterns, understanding implications.

It's super practical.

OK, so if we want to make sense of data,

step one seems to be getting organized.

The book uses the U .S.

Census Bureau's American Community Survey, the ACS.

Great example.

Think of their big data labels.

Each row represents an individual, an individual.

So like a person or it can be a person.

But here for the ACS, the individuals are the households they survey.

It's the object being described.

Got it.

Objects and the columns.

Those hold the variables, attributes that can change from one individual, one household to the next.

Like what region they're in or how much money they make.

Exactly.

Region, income, number of people.

Those are all variables for the households.

Now, this is where things get specific.

We need to know if a variable is categorical or quantitative.

That sounds important.

It's fundamental.

A categorical variable basically assigns a label.

It puts individuals into groups.

So like region, northeast, south, midwest, west, those are categories.

Right.

Or survey response mode, online, paper, phone, or even internet access.

Yes or no.

Labels, groups.

OK, so what's quantitative then?

Quantitative variables take numerical values where the numbers actually mean something as measurements or counts.

So number of people in the household, that's a count.

Or household income in dollars, that's a measurement.

They have units and doing arithmetic with them like finding an average makes sense.

But here's a tricky bit.

The book highlights not all numbers are quantitative.

Oh, absolutely.

Think about zip codes.

Right.

It's a number.

Yeah.

But you wouldn't calculate the average zip code of a town.

It's just a label for an area.

Making it categorical.

Context is everything.

The book also mentions time in dwelling.

If it's recorded in intervals like two to four years, that's categorical.

But if it was recorded as, say, 3 .5 exact years.

Then it would be quantitative.

It depends entirely on how the data is recorded and what the number represents.

Let's test this out.

The census at school data for 10 Canadian students.

The individuals are clearly the students.

Yep, 10 students.

Now let's classify some variables.

Province they live in.

Categorical.

Put some in a provincial group.

Gender.

Handedness.

Preferred communication method.

All categorical.

Labels, groups.

OK, what about number of languages spoken?

Quantitative.

It's a count?

Height in centimeters.

Risk circumference in millimeters.

Both quantitative measurements.

And speaking of that risk circumference, did you notice one value was 65 millimeter?

Seems kind of small for a student, right?

Yeah, that jumps out.

Like really small.

Could be a typo.

Or just a very unique student.

Exactly.

It's a reminder to always look at your data with a critical eye.

Does it make sense?

Check for those potentially weird values.

And for everyone listening, especially AP students, nailing this distinction between categorical and quantitative.

It's non -negotiable.

It dictates the graphs you use, the calculations you perform.

Get it wrong.

And, well, your whole analysis can be flawed from the start.

It's an absolute must -know for the exam.

OK, so we have individuals, variables, what's next?

The book talks about the distribution.

Right, the distribution of a variable just tells you two things.

What values the variable takes and how often it takes those values.

So for those Canadian students and their preferred communication.

You could imagine a simple bar graph.

The categories, cell phone, text, et cetera, are on the bottom axis, the horizontal one.

And the height of the bar shows how many students chose each one.

Like four students for text messaging?

Yep.

That shows the distribution.

Or for a quantitative variable, like number of languages spoken.

You might use a dot plot.

A number line showing one language, two languages, three.

And you stack dots above each number to show how many students reported that many languages.

Maybe six dots above one, two dots above two, and so on.

That's its distribution.

So what's the general game plan for analyzing data?

It's usually a two -step process.

First, look at each variable by itself.

Understand its own distribution.

Then what?

Then you start looking for relationships among the variables.

How do they connect?

And always start with?

Graphs.

Always.

Always start with graphs.

Visualize the data first.

Then you can add numerical summaries to get more precise.

This all feels like describing the data we have.

But statistics often goes beyond that, right?

Absolutely.

That's where inference comes in.

The book gives a little preview with that hiring discrimination activity.

You take data from a sample, maybe run an experiment or simulation, and then you ask, could these results have just happened by random chance?

What are the chances?

Exactly.

That question is the heart of probability in statistical inference.

It lets us make conclusions about a larger population or broader context, not just the specific data points we collected.

It's a huge part of statistics.

OK, fascinating.

Let's dig into section 1 .1 then.

Analyzing categorical data.

We just talked about summarizing preferred communication for those students.

Right.

We can use tables.

A frequency table just shows the counts for each category.

Like cell phone hand, two students, text messaging, four students.

And a relative frequency table shows the proportion or percent.

So cell phone, 2 out of 10 is 20%.

Text messaging, 4 out of 10 is 40%.

Useful for comparison.

And for displaying this, you mentioned bar graphs.

Yep, bar graphs are perfect.

Yeah.

Categories on the horizontal axis, frequency, a relative frequency on the vertical.

Make sure the bars have equal width, and crucially, gaps between them.

Why the gaps?

The gaps emphasize that the categories are distinct separate groups.

It's not a continuous scale like in some other graphs we'll see.

What about pie charts?

Pie charts are good when you want to emphasize how each category relates to the whole, like parts of 100%.

Each slice's area is proportional to the relative frequency.

Like for the what's on the radio example.

You could see what slice of the pie country music takes up compared to news talk.

Exactly.

Though pie charts can be harder to make accurately by hand, and you need all the categories to make the whole pie.

Bar graphs are often more flexible and easier to compare heights accurately.

Let's walk through that radio station example.

Lots of formats listed.

Adult, contemporary, country, news talk, with counts for each.

If you were making a frequency bar graph, you'd list those formats along the bottom.

Then your vertical axis would need a scale, maybe from zero up to 5 ,000, since some categories have thousands of stations.

And then draw bars above each format name, matching the height to the count.

Precisely.

Equal width bars, gaps in between.

And what would that graph tell you?

Well, looking at the data, you'd see the tallest bars are for other news talk information and religious.

Those are the most common formats.

And the shortest.

Looks like oldies, Spanish language, and contemporary hits, way fewer stations.

The graph makes that comparison instant.

But graphs can be dangerous tools if used poorly.

Oh yeah.

The temptation to make them flashy can lead you astray.

The book warns about good and bad graphs.

Big time.

Beware the pictograph.

Remember that IMAC buyer's example, using little pictures instead of bars.

Right, where they made the IMAC picture taller and wider for the group with more buyers.

Exactly.

If you make an image twice as tall and twice as wide, its area becomes four times larger.

Our eyes react to the area, so it massively distorts the comparison.

A small difference looks huge.

So stick to simple bars of equal width.

What other traps are there?

Watch those scales.

Especially the vertical axis on bar graphs.

That IMAC example also showed a graph where the vertical axis started at, say, 10 % instead of 0%.

So it chops off the bottom of the bars?

And makes the differences at the top look way bigger than they really are.

It exaggerates everything.

Key takeaway then.

For honest bar graphs,

keep bar widths consistent and always start that vertical axis at zero.

Absolutely crucial for accurate perception.

Okay, so that's handling one categorical variable.

What if we want to look at the relationship between two categorical variables?

That's where two -way tables come in.

They organize data based on two categories simultaneously.

Like the Yellowstone survey example.

Snowmobile use versus whether someone is in an environmental club.

Right, you'd have rows for snowmobile use, like never, rent, own, and columns for environmental club membership.

No, yes.

The cells in the table show the count for each combination, like how many people own snowmobiles and are in a club.

And the table usually includes totals for each row and column, called marginal totals.

Yep, and from this table, we can calculate different kinds of percentages or relative frequencies.

First is marginal relative frequency.

Marginal, like in the margins.

Exactly, you use the row or column totals divided by the overall total.

It tells you the percentage for one variable ignoring the other.

So for Yellowstone, what percent of all surveyed people were environmental club members?

You'd use the column total for yes members.

Right, comes out to about 20%.

Or the percentage who never use a snowmobile using the row total for never, which was 43 .1%.

Okay, what's joint relative frequency then?

Joint means looking at the intersection of two categories.

You take the count in one specific cell inside the table and divide by the overall total.

So the percentage who are both environmental club members and own a snowmobile, you find that specific cell.

Which was only 1 .0 % of the total sample.

It tells you how common that specific combination is.

But the really revealing one seems to be conditional relative frequency.

Yes, this is key for exploring relationships.

You're calculating a percentage within a specific condition or group.

Like, given that someone is an environmental club member, what percentage of them own a snowmobile?

You're only looking at the yes column now.

Precisely.

Or take the Titanic data.

What percentage of first class passengers survived?

Your condition on first class.

And calculating that percentage, 61 .8%, and comparing it to the percentage of third class passengers who survived.

That highlights a potential relationship between class and survival.

Exactly, conditional distributions help you compare the groups.

Back to Yellowstone.

What percentage of environmental club members never used a snowmobile?

Okay, look only at the yes club members, find the never used count within that group, and divide by the total number of club members.

Right, and it's 69 .5%.

Now compare that to the non -members.

What percentage of non -members never used one?

Do the same calculation, but looking only at the no club member group.

That was 36 .4%.

See the difference.

69 .5%, 36 .4%.

That comparison strongly suggests a connection.

And a really important AP exam tip here.

Let me guess, use percentages.

Always use relative frequencies, percentages, or proportions when comparing groups, especially if the groups have different sizes.

Counts can be really misleading.

Comparing 59 .5 % to 36 .4 % is much more informative than just comparing the raw counts.

Makes sense.

How do we visualize these relationships between two categorical variables?

Two main ways, side -by -side bar graphs and segmented bar graphs.

You basically make separate bar graphs for one variable, like snowmobile use, never, rent, own, but then you put the bars for the different groups, club member, no versus yes, next to each other for comparison.

So for never used, you'd have a bar for non -members right next to a bar for members.

Allows easy visual comparison of heights.

Exactly, and segmented bar graphs take a different approach.

You create one bar for each group, like one bar for non -members, one for members, making each bar total 100%.

And then you divide or segment that bar based on the proportions of the other variable.

Precisely, so the non -member bar might be segmented into percentages for never, rent, and own.

Same for the member bar.

You then compare the segments across the bars.

Like in the Titanic example graph, you'd have bars for first, second, and third class, each adding up to 100 % segment into survived and died portions.

And you'd visually compare, say, the size of the survived segment across the three class bars.

It makes the difference in survival rates really stand out.

So if those segments look different across the groups.

Then there's likely an association between the two variables.

Knowing the value of one variable helps you predict the value of the other.

In the Yellowstone example, the segmented bars for members and non -members looked quite different, suggesting an association between club membership and snowmobile use.

If there were no association, the segments in each bar would be roughly the same height, the same proportions.

Here comes the big flashing warning sign.

Say it loud.

Association does not imply causation.

Just because environmental club membership and snowmobile use are associated doesn't mean one causes the other.

Absolutely not.

There could be other factors involved.

The book uses that example about overweight people sometimes having lower death rates.

Seems weird, right?

Until you consider a lurking variable like smoking.

Smokers might be thinner on average, but have much higher death rates.

Smoking could be the real driver, creating a misleading association between weight and death when you don't account for it.

So the mantra is, beware other variables.

Association is a hint, not proof of cause and effect.

Always keep that in mind.

Okay, that covers categorical data really well.

Let's shift now to section 1 .2, displaying quantitative data with graphs.

Right, moving from categories to numbers that measure or count.

One of the simplest starting points is the dot plot.

Dot plot, sounds straightforward.

It is, you just draw a number line that covers the range of your data.

Then for each data value, you place a dot above its location on the number line.

And if values repeat.

You stack the dots vertically, it's quick, easy, and shows the shape of the distribution pretty well, especially for smaller data sets.

Like the U .S.

women's soccer team goals per game.

Five, five, one, 10, one, two.

You draw a number line from maybe zero to 10.

Put a dot above five, another dot above five, a dot above one, a dot way out at 10, another dot above one, one above two, stacking them up.

And you'd quickly see where the goals tended to cluster, maybe around one, two goals.

And you'd see those high scoring games at nine and 10 standing out.

Exactly, it gives you an immediate visual summary.

Now, when we look at these graphs, we need language to describe what we see.

First up,

shape.

Shape, okay, what are we looking for?

Look for the overall pattern.

Are there major peaks?

Where are the clusters of data?

Are there any obvious gaps?

And then is it symmetric or skewed?

Symmetric means like a mirror image.

Roughly, yes.

The right side looks pretty much like the left side.

If it's not symmetric, it might be skewed.

The skiing analogy.

Skewed to the right means the long tail, the gentle slope stretches out to the right towards the higher values.

Perfect, and skewed to the left means the long tail stretches to the left towards the lower values.

So if a class takes a test, and most students do really well, but a few score very low, the distribution of scores would likely be.

Skewed to the left, the main cluster is high, the tail points low.

What about rolling a die many times?

The number of times each face, one to six, comes up.

That should be roughly symmetric and also relatively flat or uniform because each outcome is equally likely.

No major peak, frequency is about the same.

Can distributions have multiple peaks?

Definitely.

We call a distribution with one main peak unimodal.

Two peaks is bimodal, more than two is multimodal.

Like the old faithful geyser example.

Its eruption durations have that dot plot with two distinct humps.

Exactly.

One cluster around two minutes, another around 4 .5 minutes.

That's clearly bimodal.

You'd want to mention both peaks when describing it.

Do some types of data tend to have certain shapes?

Often, yes.

Biological measurements like human height often tend to be roughly symmetric and unimodal.

Things like income or house prices, though, are very often skewed to the right.

Why skewed right?

Because there's a lower limit, usually zero, but no strict upper limit, and a few very high earners or very expensive houses pull that tail out to the right.

Shape is one part.

When we describe the overall distribution, the book suggests an acronym.

Yes, SOCV.

It stands for shape, outliers, center, variability.

Some people say SOCS, using spread instead of variability.

Either works.

Describe the shape.

Note any obvious outliers, values that stand far apart.

Estimate the center, like a typical value, and describe the variability or how spread out the data is.

Exactly.

Look for the main pattern, shape, center, variability, and also any striking deviations from that pattern, outliers.

And the AP exam tip.

Always, always, always include context.

Don't just say the center is five.

Say the center of the goal distribution is about five goals.

Use the variable name in units.

For now, when we mentioned center and variability in SOCV.

We'll usually estimate the median for center and use the minimum and maximum values, or the range, to describe variability.

More precise measures come later.

Let's try SOCV on that Toyota 4Runner fuel economy dot plot.

Okay, shape looks roughly symmetric, maybe slightly skewed right, with a single peak around 22 .4 millipede gen.

There seem to be some gaps.

Outliers.

There are a couple of dots that look a bit separate, maybe around 21 .5 and 23 .3 millipede gen.

Potential outlier.

Center, the middle value, the median, seems to be right at that peak, 22 .4 millipede gen.

And variability.

The data ranges from a minimum of 21 .5 millipede gems to a maximum of 23 .3 millipede gen.

Perfect.

That's a concise SOCV description, with context.

What about comparing distributions?

Like the household size data for the UK versus South Africa.

Comparison is huge in statistics.

When you compare two or more quantitative distributions, you must compare their SOCV.

Using comparative language.

Yes, use words like greater than, less than, similar to, more variable than.

Don't just list SOCV for each group separately.

Explicitly compare them.

So for UK versus South Africa household size,

UK is roughly symmetric, South Africa is skewed right.

Outliers.

UK seems to have none, but South Africa has potential high outliers around 15 and 26 people.

Center.

South Africa's center, median around six people, appears larger than the UK's center, median around four people.

Variability.

South Africa's household sizes are much more variable, ranging from three to 26 than the UK's, ranging from two to six.

Remember context.

Household size.

Got it.

Compare SOCV using comparative words and always include context.

Beyond dot plots, what else?

Stem plots.

Stem plots or stem and leaf plots are another good option, especially for smaller data sets.

They separate each value into a stem, all digits except the last, and a leaf, the final digit.

So for a value like 23 .4, the stem is 23 and the leaf is four.

Usually yes.

Or for a value like 157, stem 15, leaf seven.

You list the stems vertically, then write each leaf next to its stem.

Key steps.

Make the stems, smallest to top, include all stems in the range, even if they have no leaves.

Add the leaves, usually unordered first, then ordered, and always include a key.

The key is vital.

It tells someone how to read your plot, like 22 .4 represents a fuel economy of 22 .4 mJ.

Sometimes a stem plot looks too bunched up.

Yeah, if too many leaves are on just a few stems, you can split stems.

For example, one stem for leaves 04 and another identical stem for leaves 59.

It stretches the plot vertically, sometimes revealing shape or gaps better.

And for comparing two groups.

Use a back -to -back stem plot.

Common stems in the middle, leaves for one group going left, leaves for the other group going right.

Great for comparing shapes and centers visually,

like resting pulse versus after exercise pulse.

Okay.

But what if you have a lot of data?

Dot plots and stem plots get messy.

That's where histograms shine.

They group nearby values into intervals or bins of equal width.

So instead of a dot for every single value, you count how many values fall into each interval.

Exactly.

Then you draw bars for each interval, where the height of the bar represents the frequency count or relative frequency percent of values in that interval.

And unlike bar graphs for categorical data, the bars in a histogram touch.

There are no gaps between bars unless an interval truly has zero observations because the horizontal axis represents a continuous numerical scale.

Making one.

Choose equal width intervals, usually at least five, make a frequency table, draw and label your axis.

Variable with units on horizontal frequency, relative frequency on vertical,

scale the axes and draw the touching bars.

The choice of interval width can matter though.

Too wide and you lose detail might hide peaks.

Too narrow and it might look too noisy, less like an overall shape.

Like the state sales tax example, using 1 % intervals showed a peak around six, 7%, but using 0 .5 % intervals might show more detail within that peak.

Right, that's a bit of an art sometimes.

And remember that AP exam tip again?

Label and scale axis, can't stress it enough.

Now some common histogram pitfalls.

First, don't confuse them with bar graphs.

Right, histograms equals quantitative data, numerical intervals on horizontal axis, bars touch, bar graphs categorical data, categories on horizontal axis, bars have gaps.

Second, when comparing groups with different numbers of observations, use relative frequency histograms.

Comparing raw counts can be misleading if one group is much larger than the other, like comparing word lengths in a 400 word passage versus a 100 word passage.

Use percentages, not counts for a fair comparison.

And finally, use histograms appropriately.

Don't make one just cause.

A bar graph for quantitative data like individual student heights doesn't make sense.

A histogram showing the distribution of heights does.

Okay, that covers graphing quantitative data.

Now, section 1 .3, describing it with numbers.

Graphs give us the visual, numbers give us precision.

Exactly, we need ways to measure the center and the variability numerically.

Let's start with center, the most common measure.

The mean or average, symbol X bar.

Calculation is simple, add up all the values, divide by how many values there are.

For those 20 soccer games, add up all the goals, divide by 20.

Which came out to X steeple 3 .15 goals per game.

But what's the big caution with the mean?

It's sensitive, pulled by extreme values, those outliers, it's not resistant.

Right, if we took out those nine and 10 goal games, the mean would drop quite a bit, down to 2 .44 goals.

Outliers drag the mean towards them.

Think of it as the balance point.

So if we need a resistant measure of center, we use the median,

the midpoint of the data.

How to find it, order the data smallest to largest.

If N, the number of values is odd, it's the single middle value.

If N is even, it's the average of the two middle values.

For the 25 Toyota fuel economies, odd number, the median was the 13th value when ordered, 22 .4 mpg.

For the 20 soccer games, even number, we average the 10th and 11th ordered values, which turned out to be two goals.

So how do mean and median relate to shape?

If the distribution is roughly symmetric with no major outliers, the mean and median will be close together, like the fuel economy data.

If it's skewed right.

The mean gets pulled higher than the median by that long right tail.

Think salaries means salary is usually much higher than median salary because of a few superstars.

And skewed left?

Mean gets pulled lower than the median.

The median being resistant stays put in the middle, while the mean chases the tail.

So for skewed data or data without outliers, the median often gives a better sense of the typical value.

Generally, yes.

Choose your measure of center wisely based on the distribution.

Okay, center covered.

Now measuring variability.

How spread out is the data?

Simplest measure.

The range, maximum value minus minimum value.

Soccer goals.

Max 10, min one, range equals nine goals.

Easy.

Easy, but not great.

Why?

It only uses two values, the extremes, so it's totally not resistant.

One outlier drastically changes the range.

Yeah.

And it tells you nothing about the spread of the data in between the extremes.

Like those two nail machines in the book, same range, very different consistency.

We need something better, especially if we're using the mean as our center.

Which leads to standard deviation.

Symbol S, sub X, SX.

Standard deviation measures the typical or average distance of the observations from their mean.

Think of it as, on average, how far do data points stray from the mean?

Calculating, it looks a bit involved.

Conceptually.

Find the mean.

Calculate the deviation for each value.

Value mean.

Square those deviations, so positives and negatives don't cancel out.

Sum the squared deviations, divide by N1, this is the variance.

Take the square root.

Shoot, why N1?

That's related to making it an unbiased estimator for the population variance, a concept you'll likely explore more later.

For now, just know the formula uses N1 in the denominator for sample standard deviation.

Example, the how many friends data for 11 students.

Mean was three friends.

Standard deviation came out to XX, 1 .34 friends.

Interpretation is key.

The number of close friends these students have typically varies by about 1 .34 friends from the mean of three friends.

Properties of standard deviation.

It's always greater than or equal to zero.

Zero only if all data values are identical.

Larger sex means more variation.

It is not resistant outliers inflated.

It's typically paired with the mean.

And variance, fact squared, is just standard deviation before the square root.

Yes.

Less interpretable because its units are squared, like goals squared.

But it's important theoretically.

Standard deviation brings it back to the original units.

Okay, so if the mean and standard deviation are resistant, what's the resistant measure of variability paired with the median?

The interquartile range or IQR.

Interquartile range, okay, break that down.

First, find the quartiles.

They divide the ordered data into four roughly equal parts.

Q1, the first quartile, is the median of the lower half of the data.

Values below the overall median.

And Q3, the third quartile, is the median of the upper half.

Values above the overall median.

The overall median is sometimes called Q2.

Right.

Then the IQR is simply Q3 minus Q1.

It's the range span by the middle 50 % of the data.

Middle 50%, so it ignores the lowest 25 % and highest 25%.

Exactly, which is why it's resistant to outliers and skewness.

Those extreme values don't affect Q1 and Q3 much, if at all.

Like the boys in their shoes example, finding Q1, Q3, and then IQR gives a measure of spread for the middle chunk of shoe ownership.

And the IQR leads directly to a formal rule for identifying outliers.

The 1 .5 IQR rule, how does that work?

You calculate two fences, a lower fence, Q1, 1 .5 IQR, an upper fence, Q3 plus 1 .5 IQR.

Okay, Q1 minus 1 .5 IQR is Q3 plus 1 .5 IQRs.

Any data value falling below the lower fence or above the upper fence is flagged as a potential outlier.

Let's try it, Toyota fuel economy.

Q1, 22 .2, Q3, 22 .6.

So IQR with 0 .4 milliliter.

Lower fence, 22 .2, paint 1 .5, 0 .4 equals 22 .2 .6.

So it's 21 .6 milliliter, upper fence.

22 .6 plus 1 .5, 0 .6 plus 0 .6 is 23 .2 milliliter.

So any value below 21 .6 or above 23 .2 is an outlier.

That flags 21 .5 and 23 .3 milliliter, just like we suspected from the dot plot.

Exactly, and for the soccer goals, Q1, 1, Q3, 4 .5, IQR, 3 .5.

Lower fence, 1, 1 .5, 3 .5, 1 .5 .25 plus 5 .25 plus 9 .75 goals.

So the 10 goal game is an outlier, 9 .75, but the nine goal game is not.

The rule gives a precise threshold.

Why bother identifying outliers?

Several reasons.

Could be a data error that needs correcting, could be a genuinely unique or remarkable observation worth investigating.

And importantly, outliers can heavily distort non -resistant measures like the mean and standard deviation.

Identifying them helps you decide which summaries are most appropriate.

And for the AP exam.

Be ready to state the 1 .5x IQR rule and use it to identify outliers, showing your calculations.

And we have media dev and median IQR.

How do we visualize the median IQR picture that leads to box plots?

Right, box plots are box and whisker plots.

They are a graphical display of the five number summary.

Which is?

Minimum value, Q1, median, Q3, maximum value.

So a box plot shows all five, how?

Pretty cleverly.

You draw on a number line, then draw a box extending from Q1 to Q3.

Draw a line inside the box marking the medium.

Okay, box covers the middle 50%, median marked inside.

What about the rest?

The whiskers.

Here's where outliers matter.

First, identify any outliers using the 1 .5x IQR rule.

Then draw a whisker, a line, from Q1 down to the smallest data value that is not an outlier.

Ah, not necessarily to the minimum if the minimum is an outlier.

Correct, and draw another whisker from Q3 up to the largest data value that is not an outlier.

The outliers themselves.

Mark them individually with a special symbol like an asterisk or a dot, beyond the whiskers.

So the box plot gives you the five number summary visually and explicitly shows outliers.

Like for the forerunner data, the box plot would show the box from 22 .2 to 22 .6, median at 22 .4, whiskers extending to the lowest size non -outliers, and asterisks for 21 .5 and 23 .3.

Exactly, the pumpkin weight example also shows this while a box plot clearly displaying the right skew, medium closer to Q1, and identifying two high outliers.

For skewed data with outliers, median and IQR, shown by the box plot, are often better summaries than mean and standard deviation.

Do box plots have downsides?

Yes,

they hide certain features.

You can't see individual values within the box or whiskers.

You can't easily see gaps or clusters.

And importantly, you can't see if a distribution is bimodal.

Like old faithful, the box plot wouldn't show those two distinct peaks at all.

Correct, it might look symmetric even if the underlying distribution is strongly bimodal.

So always good to pair a box plot with a histogram or stem plot if possible.

But where box plots really excel is.

Comparing distributions across different groups, placing multiple box plots side by side on the same scale is incredibly effective.

Like the Apple versus Samsung tablet ratings example, parallel box plots let you compare everything at a glance.

Let's do it, shape.

Both look skewed left, medium closer to Q3, outliers.

Apple has two low ones, Samsung none,

center.

Apple's median, 84, is slightly higher than Samsung's 83.

You can even see that Apple's Q1 is near Samsung's median, meaning about 75 % of Apple ratings are above Samsung's typical rating.

Wow, that's a powerful comparison, variability.

Samsung's IQR, the box length, is much wider IQR 11 than Apple's IQR 3.

Samsung's ratings are far more variable in the middle 50%.

And since they're skewness and outliers, comparing medians and IQRs is the way to go here.

Definitely, box plots make that comparison very efficient.

One final AP exam tip for this section, maybe the whole chapter.

Use your statistical terms carefully and correctly.

Don't say mean when you calculated the median.

Don't call the IQR a range or describe skewness incorrectly.

Precision and language is critical on the exam.

Excellent point.

So let's recap this whirlwind tour of chapter one.

We've laid some really crucial groundwork.

Classifying variables, categorical versus quantitative.

Displaying distributions using various graphs, bar charts, pie charts, dot plots, stem plots, histograms.

And describing those distributions using SOCV shape, outliers, center, variability, plus calculating numerical summaries like mean, median, standard deviation, and IQR and knowing when to use which.

And finally, using box plots and the 1 .5 IQR rule for outliers.

It's the fundamental toolkit.

These concepts allow you to start making sense of the data that's constantly thrown at you to move beyond just looking at numbers to actually understanding what they might mean.

So here's something to think about as you go about your day.

You've just got this toolkit.

Look at some data you encounter, a news report, stats about your favorite sports team, anything.

Can you identify the individuals and variables?

Can you tell if they're categorical or quantitative?

Could you sketch a quick graph in your head or calculate a simple summary?

What story might that data be trying to tell you now that you have some tools to listen?

The story is out there, waiting for you to discover it.

Thanks for joining us on the Deep Dive.