Chapter 2: Modeling Distributions of Data

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, Deep Divers.

Have you ever gotten a test score back or maybe, you know, a health report and just found yourself wondering, is this good?

Is it bad?

Like say, Emily scores a 43 out of 50 on her stats test.

On its own, that number doesn't really tell you much, does it?

No, not really.

Is 43 points like a win or does it mean she needs to hit the books more?

That's the perfect question to kick us off because, well, like you said, the answer depends entirely on how Emily's score stacks up against the distribution of all the other scores in our class.

Right.

A raw number alone can be, well, incredibly misleading sometimes.

And that's exactly our mission today.

This deep dive is all about giving you the tools to understand where any individual data point truly stands within its group.

We're drawing directly from the practice of statistics, sixth edition, aiming to arm you with practical know -how for your AP statistics journey.

We'll explore these really powerful methods for describing location and data distributions

and also how to use models to make sense of it all.

We're gonna help you look beyond just the surface number to really see the relative position and the significance of any data point you encounter.

It's such a fundamental skill in stats and it'll serve you well as you navigate through more complex data sets later on.

Absolutely.

We'll be uncovering the secrets of percentiles, Z scores and those really fascinating normal distributions.

All while giving you insights that are super practical for real world analysis and, of course.

For the AP stats exam.

Okay, let's untack this.

Okay, let's circle back to Emily's test.

She scored 43 out of 50 in Mr.

Taber's class.

Now there were 25 scores in total.

And if you were to visualize those scores, maybe on a dot plot, you'd actually see the distribution was skewed a bit to the left.

There were some lower scores pulling that tail down.

So not perfectly symmetric.

No, and Emily's 43 was visibly above the class average, sure, but how much better was it, you know, relatively speaking?

That's the crucial point.

The raw score, 43,

it just isn't enough context on its own.

What truly matters is Emily's relative position within that specific group.

Did she score better than most or just like a few?

Exactly.

And that's where percentiles come in, right?

Precisely.

An individual's percentile is simply the percentage of values in a distribution that are less than that individual specific data value.

Okay, so for Emily, if we counted up the scores, turns out 20 of the 25 scores were below her 43.

So 20 out of 25, that's 80%.

80 % of the scores were less than hers.

So Emily is at the 80th percentile.

And quick language point here, it's important to remember you are at a percentile, not in one.

It's like a specific rank or position.

Ah, okay, good distinction.

So imagine Jacob in the same class, he scored an 18.

Ouch.

Yeah, only one score out of the 25 was less than his 18.

So his percentile would be?

One out of 25, that's 4%.

He's at the fourth percentile.

Quite a different story from Emily's 80th.

Definitely.

Now let's think about Maria.

Her test score was at the 48th percentile.

So what does that tell us?

It means 48 % of the students scored lower than Maria.

And with 25 students, .48 times 25 is 12.

So 12 students scored lower.

Exactly.

Her actual score happened to be a 38.

And interestingly, if other students also scored 38.

They'd also be at the 48th percentile.

Correct, because the number of scores less than 38 is still 12.

That calculation doesn't change.

Gotcha.

And a quick side note, but it's important.

A high percentile isn't always a good thing, is it?

Oh, definitely not.

Good point.

Like if your cholesterol level is at the 90th percentile for your age group.

Yeah, that probably means 90 % of people your age have lower cholesterol.

Not exactly what you want to hear from your doctor.

Right, context matters.

It's also worth connecting this back to some familiar measures.

The median, for instance, the middle value, that's roughly the 50th percentile.

Makes sense, half above, half below.

And the first quartile, Q1, that's approximately the 25th percentile.

And the third quartile, Q3, is roughly the 75th percentile.

They mark out those quarter points in the data.

Okay, so let's check our understanding maybe for you listening.

If Mark earned a raw score of 39, and that placed him at the 68th percentile, what does that really mean?

It means he scored higher than about 68 % of the students who took that test.

And what if a doctor tells you your daughter is at the 87th percentile for weight and the 67th for height?

Well, it means 87 % of girls her age, way less than she does.

And 67 % of girls her age are shorter than she is.

It tells you her size relative to her peers.

Okay, now visualizing all these percentiles,

that can be tricky.

It can, but there's a great tool for that.

Cumulative relative frequency graphs.

Sometimes they're called OGEVs, O -G -I -V -E -S.

OGEVs, right, fancy name.

A little bit.

But they're basically graphs that display percentiles visually.

They give you a continuous picture of how the data accumulates.

So how do you build one?

Let's imagine the ages of the first 45 US presidents when they took office.

We'd start by grouping their ages, maybe in five -year bins.

Then for each bin, we figure out the percentage of presidents inaugurated at or below the upper age limit of that bin.

So maybe no presidents took office before age 40.

The graph starts at 0 % cumulative frequency at age 40.

Right, and then maybe it rises to say 4 .4 % by age 45, meaning just over 4 % were 45 or younger when inaugurated.

And here's where it gets really interesting, I think.

The slope of this OGEV tells you where the data is concentrated.

How so?

Well, if the graph is steep in a certain age range, say between 50 and 59,

that means a large percentage of presidents fall into that age group.

Lots of data points there.

If the graph is flatter, like maybe in the early 40s or late 60s.

Fewer presidents were inaugurated at those ages.

Exactly, fewer data points, so the cumulative percentage grows more slowly.

The graph always climbs smoothly from 0 % up to 100 % by the time you reach the maximum observed age.

Okay, and once you have this graph, interpreting it seems pretty useful.

Let's take Barack Obama inaugurated at age 47.

You'd find 47 on the horizontal axis age, trace vertically up to the curve, and then trace horizontally over to the vertical axis, cumulative relative frequency.

You'd estimate maybe it's around the 11th percentile.

Yeah, sounds about right from looking at typical data.

So that tells you he was fairly young, relatively speaking, but maybe not extremely so.

About 11 % were younger.

And you can go the other way.

What if you wanna find the age corresponding to the 65th percentile?

You'd start at 65 % on the vertical axis this time, trace horizontally to the curve.

And then drop straight down to the age axis.

Right, and you might land on, say, roughly 58 years old.

Meaning?

Meaning about 65 % of those first 45 presidents were younger than 58 when they took office.

Wow, okay.

That's pretty clear.

You could use this for lots of things, right?

Like phone call lengths?

Absolutely.

You could quickly estimate the median call length, 50th percentile, or the IQR, by finding the ages at the 25th and 75th percentiles and subtracting.

Very visual.

Okay, percentiles and OGEVs give us a good sense of relative standing.

But sometimes we need something that also accounts for the spread or variability in the data.

Yes, exactly.

And that brings us to z -scores.

Z -scores, heard of those.

They're fundamental.

A z -score, or a standardized score, tells us precisely how many standard deviations a specific value falls away from the mean of its distribution.

And importantly, in which direction, above or below.

And the formula is pretty straightforward, isn't it?

It's just z -value mean standard deviation.

That's the one.

Simple but powerful.

So let's revisit Emily's test score.

She got a 43.

The class mean was 35 .44, and the standard deviation was 8 .77.

Okay.

Plugging that in.

43, 35 .44, 8 .77.

That comes out to about .86.

So what does that .86 mean?

It means Emily's score is .86 standard deviations above the class average.

It standardizes her performance.

Perfect.

And Jacob, with his score of 18, same mean, same standard deviation.

His calculation would be 18, 35 .44, 8 .77.

That's roughly negative 1 .99.

So Jacob's score is almost two standard deviations below the mean.

The negative sign tells us it's below average.

The 1 .99 tells us how far below, using the standard deviation as our measuring stick.

And you can even work backwards, right, if you know someone's z -score.

Like Tamika had a z -score of 0 .292.

Yep.

To find her actual score, you just rearrange the formula.

Score equals mean plus z -score standard deviation.

So 35 .44 plus .292, 8 .77.

That gives you 38.

Her raw score was 38.

Exactly.

But the real magic of z -scores, I think, is their ability to let us compare observations from completely different distributions.

Oh, right, like comparing apples and oranges.

Sort of.

Imagine Jordan, a nine -year -old girl who's 55 inches tall.

Let's say for her age group, that gives her a z -score of plus one.

She's one standard deviation taller than the average nine -year -old girl.

Now consider Zane, an 11 -year -old boy who is 58 inches tall.

Let's say for his age group, that height corresponds to a z -score of plus 0 .5.

So Zane is actually taller in inches, 58 versus 55.

Correct, but who is taller relative to their peers?

Jordan is.

Her z -score of one is higher than Zane's z -score of 0 .5.

Exactly.

Even though Zane is objectively taller, Jordan is more unusually tall for her specific comparison group.

Z -scores put everyone on that same standardized scale, making these kinds of comparisons totally valid and really insightful.

That's really useful.

Okay, so we've looked at individual scores, but what happens if we change all the data?

Let's talk about transforming data.

Good transition.

Yeah, sometimes we need to transform data, maybe change units, like Celsius to Fahrenheit, or correct measurements from a faulty instrument.

So what happens to the distribution's shape, its center, its spread when we do that?

Great question.

Let's break it down into two main types of transformations.

First,

adding or subtracting a constant.

Like adding five points to everyone's test score.

Exactly.

When you add or subtract the same number a to every single observation.

What changes?

Your measures of center and location, the mean, median, quartiles, min, max, they all shift up or down by that amount a.

Okay, that makes sense.

The whole distribution just slides over.

Right, but here's the key.

Measures of variability range, IQR, standard deviation, they do not change at all.

Really?

The spread stays the same?

Stays exactly the same.

Right.

And crucially, the shape of the distribution also doesn't change.

If it was skewed left before, it's still skewed left after you add five points.

So Mr.

Taber adds five points.

The dot plot just shifts right by five units, same shape, same spread, just higher scores overall.

Mean and median are five points higher.

Perfect.

Now the second type,

multiplying or dividing by constant.

Like converting scores to percentages by multiplying by two?

Yep.

When you multiply or divide every observation by the same positive number b.

What happens then?

Everything changes.

Well, almost everything.

Measures of center, location, and measures of variability all get multiplied or divided by b.

Ah, okay, so the mean doubles, the median doubles?

And the standard deviation doubles, the IQR doubles, the range doubles, the whole distribution stretches or shrinks.

But the shape?

The shape still doesn't change.

If it was skewed left, multiplying by two stretches it out, but it keeps that same fundamental skewed left shape.

So converting Celsius to Fahrenheit involves multiplying by 95 and then adding 32.

Right.

So the mean temperature in Fahrenheit would be 95 mean in Celsius plus 32.

It reflects both operations.

But the standard deviation?

The standard deviation in Fahrenheit would just be 95 standard deviation in Celsius.

The addition of 32 shifts the center, but it doesn't affect the spread.

Only the multiplication scales the spread.

This ties back to z scores, doesn't it?

It does.

Standardizing is a transformation.

You subtract the mean, that's subtracting a constant, and then you divide by the standard deviation.

That's dividing by a constant.

So when you convert an entire data set to z scores.

The new distribution will always have a mean of zero and a standard deviation of one, always.

But importantly, it's original shape, perfectly preserved.

Whatever shape you started with, that's the shape of the z score distribution.

Okay, that clarifies transformations.

Now sometimes we have so much data, right?

Like thousands of points.

Looking at individual dots becomes less useful.

Exactly.

We start looking for the overall pattern, the general flow.

And that's where we move from actual data to idealized models, specifically density curves.

Density curves.

Sounds mathematical.

It's a mathematical model, yes.

But the idea is intuitive.

It's a curve that smoothly describes the overall pattern of a distribution.

Think of it as smoothing out the bumps in a histogram.

What are the rules for these curves?

Two main things.

First, a density curve is always on or above the horizontal axis.

You can't have negative proportion.

Second, the total area underneath the entire curve is exactly equal to one or 100%.

It represents all the observations.

And the area under the curve over a specific range?

That area gives you the proportion of all observations that fall within that range.

It's a way to model probabilities or percentages for continuous data.

Let's make this concrete.

Imagine Selena's train commute time.

Maybe we look at a thousand trips and the times are kind of spread out evenly between say two and five minutes.

Roughly uniform distribution of travel times.

How would a density curve model that?

The simplest model would be a uniform density curve, basically, a flat rectangle.

A rectangle.

The base of the rectangle would go from two to five minutes on the horizontal axis, so its width is three minutes.

Since the total area must be one, the height of the rectangle has to be 13, because width height equals three, 13 equal one.

Ah, I see.

The height is set to make the total area equal one.

Exactly.

This simple rectangle smoothly models the overall pattern.

Now, if you wanted to know the proportion of times, her trip took less than four minutes.

You'd find the area under the curve between two and four minutes.

Precisely.

That's a rectangle with two from two to four and height 13.

The area is two, 13 equals 23.

So about 66 .7 % of her trips were under four minutes, according to the model.

That's it.

The curve smooths out the data's little irregularities and gives us the big picture.

So quick check.

If you had random numbers generated uniformly between zero and two, the density curve would be a rectangle from zero to two.

With height.

Width is two, so height must be 12, to make the area one.

You got it.

And then you could find the proportion of numbers between, say, 0 .5 and 1 .5, just by calculating the area of that slice of the rectangle.

Okay, cool.

Now, do these density curves have means and medians like real data?

They do.

We think about them slightly differently, though.

The median of a density curve is the equal areas point the value that splits the area under the curve exactly in half.

50 % on each side.

And the mean?

The mean, which we denote with a Greek letter, mu, mu, for models, is the balance point.

If the curve were made of solid material, the mean is where it would balance perfectly.

So for a symmetric curve, like a perfectly bell -shaped one.

The mean and median are exactly the same, right at the center point of symmetry.

But if the curve is skewed.

The mean gets pulled towards the long tail, away from the median.

So for a right skewed curve, the mean will be greater than the median.

And for a left skewed curve, the mean is pulled left.

So the mean is less than the median.

Exactly.

And just like we use mu for the mean, we use the Greek letter sigma for the standard deviation of a density curve or model.

This helps distinguish these theoretical parameters from the sample statistics, X bar and S, we calculate from actual data.

Got it.

Omen for models, X bar and S for data.

Okay, now,

you mentioned bell -shaped curves.

That leads us to, arguably, the most important family of density curves.

The normal distribution.

Also known, of course, as the bell curve.

The famous one.

Indeed.

Normal distributions are a very specific family of density curves.

They're always symmetric,

single -peaked, and have that characteristic bell shape.

And what makes them so special, besides the shape?

What's incredibly useful is that any normal distribution, no matter where it's centered or how spread out it is, is completely defined by just two numbers.

It's mean and it's standard deviation.

Just those two.

Mean and standard deviation tell you everything about that specific normal curve.

Everything.

Give me the mean and standard deviation, and I know exactly which normal curve we're talking about.

Can you estimate the standard deviation just by looking at the curve?

You can, actually.

It's related to the curvature.

Imagine skiing down the side of the bell curve.

It starts steep, then becomes less steep.

The points where that change in steepness happens, the inflection points, are exactly one standard deviation away from the mean on either side.

Huh, that's neat.

And we see things that are approximately normal all the time in real data.

Test scores, like the ITBS vocabulary scores for seventh graders in Gary, Indiana, they found those were approximately normal with the mean end of 6 .84 and a standard deviation of 1 .55.

So if you're sketching that, you draw the bell shape, center it at 6 .84, and then mark out points 1 .55 units away on either side for one sub, then another 1 .55 units for two, and so on.

Exactly.

Labeling those points corresponding to one, two, and three standard deviations away from the mean is really helpful for visualizing.

Why are these normal distributions so darn important in statistics?

Well, several reasons.

First, they're just good descriptions, good models for a lot of real world data, biological measurements, physical phenomena, test scores.

Second, they describe the results of many chance processes like sampling outcomes.

And third, maybe most importantly for AP stats, there are the foundation for many statistical inference procedures we use to draw conclusions from data.

You'll see them again and again.

And for these normal distributions, there's a really handy rule of thumb, right?

The 68, 95, 99 .7 rule, sometimes called the empirical rule.

It's a fantastic shortcut.

What does it say again?

For any normal distribution, no matter its mean or standard deviation,

approximately 68 % of the observations fall within one standard deviation of the mean.

Okay, 68 % within one SD.

Approximately 95 % of the observations fall within two standard deviations of the mean.

95 % within two SDs.

And approximately 99 .7 % of the observations fall within three standard deviations of the mean.

99 .7 % within three SDs.

Wow, that covers almost everything.

Pretty much.

It gives you a quick way to understand where most of the data lies in any normal distribution.

Let's use it.

Back to those ITVS scores.

I mean, 6 .84 SD, 1 .55.

How unusual is a score less than 3 .74?

Okay, first, where is 3 .74 relative to the mean?

The difference is 3 .74, 6 .84 divided by the SD.

One is 3 .1, 1 .55, faken's net and two.

So 3 .74 is exactly two standard deviations below the mean.

All right, so the rule says 95 % are within two SDs.

Which means 5 % are outside that range, more than two SDs away.

And since it's symmetric.

Half of that 5 % is in the lower tail.

So 2 .5 % of scores are less than 3 .74.

That's pretty unusual.

What about car stopping distances?

I mean, 155 feet, SD three feet.

What percent take more than 158 feet?

Okay, 158 feet.

That's 155 equals three feet above the mean.

Which is exactly one standard deviation above the mean.

Rule says 68 % are within one SD.

So 32 % are outside one SD, more than one SD away.

By symmetry, half of that 32 % is in the upper tail.

Exactly.

So 16 % of cars take more than 158 feet to stop.

And that also means 158 feet is at the 84th percentile.

16 % is 84%.

You got it.

See how the rule connects to percentiles.

Yeah.

But quick reminder, this rule only works for normal distributions, right?

Crucially important reminder.

Only for distributions that are approximately normal.

Okay, so what if a value isn't exactly one, two or three standard deviations away?

How do we find proportions then?

Like what percentage of ITBS scores are below six?

Six isn't exactly one or two SDs away.

Great question.

That's when we need a more general method than the 68, 95, 99 .7 rule.

We need to calculate areas under the specific normal curve.

And the key is to use standardization, convert our value to a Z score.

Ah, back to Z scores.

Z equals Z equals Z.

Always.

This transforms our specific normal problem into a problem on the standard normal distribution, in zero SD one, which we can then solve using technology or tables.

Okay, so what's the process?

Let's say we wanna find the area, the proportion for a given value.

There's a really important two -step process, especially vital for showing your work on the AP exam.

Step one, draw.

Draw the curve.

Yes, sketch a normal curve, label your access with the variable name and units,

mark the mean and the standard deviation,

clearly mark the boundary value you care about, and then critically shade the area you're trying to find.

Why is drawing so important?

It forces you to visualize the problem, shows the greater you understand what you're calculating and helps prevent simple errors.

It's a communication step.

Don't skip it on the exam.

Okay, draw and shade, step two.

Calculate, you have options here.

You can standardize your boundary value to Z scores and use a standard normal table, table A, or technology that works with Z scores.

Or, more commonly now, use technology directly, like the normal CDF function on a calculator.

This function usually lets you input the lower boundary, upper boundary, the original mean, and the original standard deviation.

Normal CDF, got it.

And another huge AP exam tip.

If you use technology like normal CDF, you must label your inputs.

Don't just write normal CDF -1006, 6 .84, 1 .55.

Write something like normal CDF lower, gain is 1 ,000, upper.

Six mean, 6 .84, SD 1 .55.

Avoid calculator speak without explanation.

Makes sense, show what the numbers mean.

Okay, let's try one.

ITBS scores, 6 .84 is 1 .55, proportion less than six.

Okay, step one, draw the curve, center it 6 .84, mark six, shade everything to the left of six.

Done, mentally.

Step two, calculate.

Using normal CDF, lower bound is effectively negative infinity.

Use a large negative number, like mega 20 ,000.

Upper bound is six, mean is 6 .84, SD is 1 .55.

Clam, what did I give you?

Calculator says about 0 .2946, so roughly 29 .5 % of scores are less than six.

Perfect, let's try another.

Car stopping distance, year 155, percent less than 160 feet.

Draw curve, center 155, mark 160, shade left, calculate normal CDF lower, nine ish to 1 ,000, upper, 160 mean, 155 SD three, Gibbs, .9525.

About 95 .3 % stop in less than 160 feet.

Excellent, what about finding the proportion above the value, say ITBS scores at least nine.

Okay, draw curve, center 6 .84, mark nine, shade right, calculate normal CDF lower, nine upper, 1 ,000 mean, 6 .84 SD, 1 .55.

That gives about .0823, so about 8 .2 % score nine or higher.

You can also find the area to the left of nine and subtract from one, right?

Oh yeah, one normal CDF lower, name is 1 ,000, upper, 90 mean, 6 .84 SD, 1 .55, should give the same three normal CDF lower, six upper, nine mean, 6 .84 SD, 1 .55.

Calculator says .6231, about 62 .3 % of scores fall in that range.

See, draw, calculate, label, you can find any area under a normal curve.

Okay, that's finding areas from values.

What about the other way around?

Finding a value from a given area or percentile.

Right, working backwards.

Sometimes you know the proportion, like the top 10 % or the 25th percentile, and you need to find the actual data value, the score, the height, et cetera, that corresponds to it.

Same first step, draw.

Absolutely.

Step one, draw the normal curve, mark the mean, shade the area corresponding to the given proportion, you deshade the leftmost .90 for the 90th percentile.

Label the unknown boundary value you're trying to find maybe with a question marker X.

Okay, step two, calculate.

Yes.

Again, technology is your friend here.

Use the inverse normal function, often called NV norm.

NV norm, what does that need?

Typically it needs the area to the left of the boundary value you're looking for, the mean and the standard deviation.

Area to the left, okay, and the AP tip.

Same deal, label your inputs,

NV norm, area, .90, mean, 6 .84, SD, 1 .55.

Show you know what each part means.

Let's try it.

Find the 90th percentile of ITBS scores on 1 .84, 1 .55.

Okay, draw curve, shade the leftmost, 90%, .90 area, we need the boundary value.

Calculate, NV norm, area, .90, mean, 6 .84, SD, 1 .55.

Calculator gives 8 .824.

So a score of about 8 .8 is needed for the 90th percentile.

Nicely done, one more.

Suppose three -year -old girls' heights are normal.

94 .5 centimeters, four centimeters.

What height are 75 % of girls taller than?

Hmm, 75 % are taller.

That means only 25 % are shorter, right?

So we're looking for the 25th percentile.

Exactly, tricky wording sometimes.

So the area to the left is .25.

Draw curve, center, 94 .5, shade leftmost, .25.

Calculate, NV norm, area, .25, mean, 94 .5, SD4.

Gives 91 .82 centimeters.

So about 75 % of three -year -old girls are taller than 91 .8 centimeters.

Wow, okay, these normal calculations are super powerful, but it all hinges on knowing the data is actually normal or at least approximately normal.

How do we check that?

Excellent and critical question, because most of the inference methods you'll learn later assume normality or rely on it.

So assessing normality is a key skill.

What are the ways?

There are a few main approaches.

First, just graphical inspection.

Make a plot of your actual data, a dot plot, a stem plot, maybe a histogram.

Then look for?

Look to see if it's roughly symmetric, single -peaked, and kind of bell -shaped.

If it's clearly skewed, like say data on number of siblings usually is skewed, right?

Or has multiple distinct peaks bimodal, then you can immediately say it's probably not normal.

Okay, so look at the picture first.

Makes sense, what else?

Second, you can check against the 68, 95, 99 .7 rule.

Calculate the actual mean and standard deviation from your sample data, X bar and S.

Then find what percentage of your actual data points fall within one, two, and three standard deviations of your sample mean.

And compare those percentages to 68%, 95%, and 99 .7%.

Exactly, if your actual percentages are reasonably close to the rules percentages, that's good evidence for approximate normality.

Like in the book's example with breakfast cereal calories, they found only 81 .8 % of values were within one standard deviation of the mean.

Right, that's way off from the expected 68%.

So even if the histogram looked vaguely bell -shaped, that calculation tells you it's not approximately normal.

But for the IQ scores example, the percentages were super close.

68 .3%, 95 .0%, and 100 % versus 99 .7%.

Yeah, those are very close fits.

That strongly supports the idea that the IQ score distribution is approximately normal.

And you mentioned a crucial AP exam tip here.

Yes,

never ever say a real data distribution is normal.

Real data is never perfectly normal.

Always use language like approximately normal or seems reasonable to assume normality.

Acknowledge that it's a model.

Okay, approximately normal, got it.

Is there a third way to check?

Yes, and it's often considered the best method, normal probability plots.

Okay, what are those?

It's a special kind of scatter plot.

For each data point in your set, you plot the actual data value on say the axis against the Z score you would expect that value to have if the distribution were perfectly normal on the X axis.

Okay,

expected Z score versus actual value.

How do you interpret that plot?

It's surprisingly simple.

If your data is approximately normal, the points on the normal probability plot will lie close to a straight line.

A straight line means normal.

Approximately normal, yes.

If the points follow along a clear curve or show some other systematic pattern deviating from a line, then the distribution is not normal.

So the IQ scores plot would look pretty straight.

Should, yes.

But the plot for number of siblings or those serial calories,

those would likely show distinct curves.

Does the way it curves tell you anything?

It can.

Often a curve that bends upwards suggests right skewness in the original data, while a curve downwards suggests left skewness.

Like the guinea pig survival times example, it's normal probability plot was clearly curved, indicating the survival times were not normally distributed, probably skewed right.

And calculators can make these plots.

Most statistical software and graphing calculators can generate normal probability plots easily, making this a very practical assessment tool.

Okay, that's a solid toolkit for checking normality.

It is.

Graphical check, 68, 95, 99 .7 rule check, and the normal probability plot.

Use them together.

So we've really covered a lot of ground here.

From wondering about Emily's single test score.

All the way to assessing if an entire data set fits the most important distribution model in statistics.

We've seen how percentiles give relative standing, how Z scores provide a standardized measure for comparison.

How transformations affect our data and the immense utility of density curves, especially the normal distribution with its rules and calculation methods.

These really are fundamental tools, aren't they, for making sense of numbers.

Absolutely indispensable.

Understanding location and distribution is foundational for pretty much everything else you'll do in statistics.

So thinking beyond the class examples,

what other everyday things could you now analyze with these tools?

You get some measurement, some observation.

How can you now truly understand its significance beyond just the raw number?

Maybe batting averages, rainfall amounts, commute times, heights of buildings, anything quantitative.

Where does that specific instance fall within the typical range?

Is it unusual?

By how much?

Yeah, thinking about it with percentiles, Z scores, maybe even seeing if it fits a normal pattern.

It gives you a much richer understanding than just the number itself.

Lots to think about.

Well, thank you so much for joining us on this deep dive into describing location and distributions.

Hope you found it helpful.

We certainly hope you feel a bit more equipped to tackle these concepts, both in your studies and as you encounter data out there in the world.

Until next time.