Chapter 6: Random Variables

Welcome to Last Minute Lecture.

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

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

For complete coverage, always consult the official text.

Okay, let's unpack this.

Have you ever, you know, scrolled through the news and seen something like 75 % of people prefer bottled water?

Right.

Or maybe you remember that old 2020 bit,

where kind of surprisingly New York City tap water actually beat the fancy bottled stuff in a blind taste test.

Exactly.

Yeah.

It really makes you think, doesn't it?

Like, how much is real preference and how much is just,

well, chance?

Can people actually tell the difference?

It's interesting how those scenarios pull us right into probability, into thinking about random outcomes.

And our source material, the practice of statistics, it kicks off chapter six with a great example.

An activity in class.

Oh, yeah.

Yeah.

Students tried to pick out bottled water from three cups.

If they were just guessing, completely guessing, the chance of being right is, well, pretty straightforward.

One in three.

Okay.

One in three.

But here's where it gets interesting, right?

In this specific class, Mr.

Hogass's class, out of 21 students, 13 got it right.

13, that's way more than half.

So that brings up the big statistical question.

If they couldn't tell the difference, if they were just guessing, how likely is it that 13 or more would get it right?

And that question, that's really the core of what we're diving into today.

This is all about getting you comfortable with random variables, which is chapter six in the practice of statistics.

Right.

We've tailored this specifically for AP statistics learners.

We want to give you a clear, you know, comprehensive, but also engaging way to understand this stuff.

Exactly.

Our goal here is to kind of demystify these concepts, show you how they actually work in the real world and importantly, how they pop up on the AP exam.

We want you walking away, feeling confident, not just with the definitions, but knowing the why and the how behind it all.

Yeah.

We'll break down the key ideas, the formulas using, you know, plain language, lots of examples, both real world and AP style.

And if there are charts or graphs in the chapter, we'll describe them so you can picture what's going on.

Plus we'll flag those common mistakes and give you those essential AP exam tips.

Sounds good.

Sounds perfect.

So let's start at the beginning.

What exactly is a random variable?

Okay.

Simplest terms.

A random variable is just a numerical outcome that results from some chance process.

Think about flipping a coin three times.

The outcomes are sequences like heads, heads, heads, heads, heads, heads, tails, et cetera.

But if we define a random variable, let's call it X as the number of heads you get, then X can only be zero, one, two, or three.

It's a number summarizing the chance result.

Got it.

So it takes the chance outcome and assigns a number to it.

And for every random variable, there's this thing called a probability distribution.

Exactly.

That's the full picture.

It could be a table, maybe a rule that lists all the possible values the variable can take on and crucially the probability for each of those values.

For the coin toss, the probability of getting zero heads, X zero is 18, one head X one is 38 and so on.

Precisely.

And the first type we need to get comfortable with is the discrete random variable.

Discreet.

The key word here is countable.

A discrete variable takes on a fixed set of values, usually whole numbers, and there are clear gaps between them like the Apgar score for newborns.

Right, that score zero, one, two, up to ten, always a whole number.

Exactly.

You can't get an Apgar score of say 7 .2.

It jumps from seven to eight.

There are gaps.

So distinct separate values means it's discrete.

And you mentioned probability distributions need to follow some rules.

Two main rules, yeah.

And they're fundamental.

First,

every single probability listed has to be between zero and one, inclusive.

Okay, no negative chances, no chances over 100%.

Makes sense.

And second, if you add up the probabilities of all the possible outcomes, the total must be exactly one, no more, no less.

If it's .99 or 1 .01, something's not right with your distribution.

So for calculations with discrete variables, it's about adding up the probabilities for the outcomes you're interested in.

Pretty much.

Let's stick with Apgar scores.

Say you wanted PXC zero, the probability of a score of zero.

If you knew the probabilities for scores one through ten, you could just add those up and subtract from one.

Using the complement rule.

Exactly.

And PX zero turns out to be really small, like .001.

Okay.

And you mentioned an AP tip here about boundary values.

Oh yeah.

This is a big one.

You have to pay really close attention to the inequality signs.

Like, is it greater than or equal to or just greater than?

So a healthy newborn might be defined as an Apgar score of seven or higher.

That's PXU seven.

Maybe that probability is say .908.

But if the question asks for the probability of a score strictly greater than seven, that's PX seven.

Now you don't include X seven.

That probability might drop to say .809.

It seems small, but on the exam, that distinction is critical.

Gotcha.

Watch those inequalities.

Now how do we visualize these histograms?

Yeah.

Histograms work great for discrete distributions.

You put the possible values of X on the horizontal axis and their probabilities on the vertical axis.

The height of each bar shows the probability of that specific value.

So for the Apgar scores, what would that look like?

It would be skewed heavily to the left.

The tallest bar would be way over on the right, probably at XZ nine because most babies score high.

The bars for low scores like zero, one, two would be tiny.

Okay.

Skewed left, peak towards the higher scores.

And what about that Jeep Tour example, Pete's Jeep Tours?

Right.

Where the variable C is the total amount collected based on passengers.

That histogram would look different.

It'd be roughly symmetric, kind of bell -shaped, but maybe flatter with a single peak, probably around the $600 mark.

It gives you a quick sense of the most likely daily earnings.

So we've got the values, their probabilities, the shape.

What about the center, the average outcome?

That's the mean or what we call the expected value.

We use the symbols EX or the Greek letter Mu, EX.

Think of it as a weighted average.

Weighted by the probabilities.

Exactly.

You take each possible value, multiply it by its probability, pi, and then sum up all these products to see a pie.

So for Pete's Jeeps, if he gets $300, 15 % of the time, $450, 25 % of the time, and so on.

You'd calculate $300 .15 plus $450 .25 plus $600 .35 plus all the way up to $900 .05.

And that gives his expected daily collection.

Right.

It comes out to $562 .50.

$562 .50.

But wait, that's not even one of the possible amounts he can collect on a single day, right?

Precisely.

And that's a key point about expected value.

It's a long run average.

If Pete runs tons and tons of tours, his average collection per tour will get really close to $562 .50.

It's not predicting any single day.

Okay.

So it's theoretical average over many repetitions.

Like the mean Apgar score being 8 .128.

No baby actually gets that score.

Exactly.

It helps with long -term planning, not forecasting one specific event.

And the AP tip for this.

Show your work.

Always.

Even if you use your calculator, like the one var stats, L1, L2 function, which is great for checking, you absolutely must show the formula with the numbers plugged in first.

Think about pie pie equals 300 .15 plus error pie pie plus error $562 .50.

Show the setup.

Okay.

Mean gives us the center.

What about the spread variability?

That's where standard deviation comes in.

Symbol is sigma.

It measures the typical deviation of outcomes from the mean.

Man.

And it's related to variance.

It's the square root of the variance.

Variance is Lx squared, sometimes written var x.

How do we calculate variance?

It's another weighted average.

This time it's the average of the square differences between each outcome and the mean.

Okay.

So outcome mean two, weighted by probability.

You got it.

The formula is Lx two pie.

For Pete's tours, you'd take 300, 562 .5 duro, 2 .15 plus 450, 562 .50, 2 .25 and so on for all values.

Sum them up.

That's your variance.

And then take the square root for the standard deviation.

Right.

For Pete, the standard deviations he seek comes out to about $162 .46.

Meaning his daily takings typically vary by about $163 from his average of $562 .50.

That's the interpretation.

Yeah.

For Apgar's scores, the standard deviation sex is about 1 .437 units.

Same AP tip applies.

Show the variance formula setup.

Absolutely.

Show the x two pie calculation before you state the variance or standard deviation, even if you used a calculator function.

Okay.

That covers discrete variables.

What about the other type?

Continuous.

Continuous random variables.

These are different because they can take any value within a given interval.

There are no gaps.

Think about generating a random number between 0 and 9.

It could be 3 .14159 or 7 .0 or 2 .5.

Any value in that range is possible.

Or maybe Selena's train commute time.

It could be 28 .5 minutes, 31 .2 minutes, 29 .753 minutes.

Infinite possibilities within a range.

Infinite possibilities.

So how do probabilities work then?

You can't list them all.

Exactly.

For a continuous variable, the probability of getting any single exact value is actually zero.

Think about it.

The chance of the commute being exactly 30 .0000 minutes is basically impossible.

Okay.

So how do we find probabilities?

We use areas under a density curve.

Yeah.

The total area under any density curve is always one, representing 100 % probability.

Probability for a continuous variable corresponds to the area under the curve over a specific interval.

Ah.

So px is between a and b.

Is the area under the curve from a to b.

Precisely.

Consider that random number y between 0 and 9.

If it's truly random, any value is equally likely.

This is a uniform distribution.

What does its density curve look like?

A simple rectangle.

It starts at y is 0 and ends at y is 9, so the base is 9 units long.

Since the total area must be 1, the height has to be 19.

Okay.

Base 9, height 19, areas 9, 19.

It wants.

Makes sense.

Now, if you want the probability that y falls between 3 and 7,

p3yA7.

You find the area of the rectangle between 3 and 7.

Exactly.

The base is now 7, 3 equals 4.

The height is still 19.

So the area and the probability is 4, 19 equals 49.

And for continuous variables, does it matter if it's a or c?

Nope.

Because the probability of any single point is 0,

p3yA7 is exactly the same as p3y7.

The boundaries don't add any area.

That's handy.

What about non -uniform curves, like the normal distribution?

Ah, yes.

The normal distribution.

Super important.

Young women's heights, for example, often follow a normal curve.

Let's say mean height is 64 inches, standard deviation is 2 .7 inches, and 64, 2 .7.

Okay.

If you want the probability a randomly chosen young woman is between, say, 68 and 70 inches tall, p68y70, you find the area under that normal curve between 68 and 70.

How do we find that area?

Z -scores.

That's the traditional way.

Convert 68 and 70 to Z -scores using value mean SD, then use a standard normal table.

Or, much more common now, use your calculator's normal CDF function.

Normal CDF, lower bound, upper bound, mean, standard deviation.

Right.

But again, the AP exam reminder,

don't just write the calculator command.

You need to show you understand what it's calculating.

Exactly.

Draw the normal curve, label the mean, 64, and standard deviation, 2 .7.

Shade the area you want between 68 and 70.

Then either show the Z -score calculations or write out the normal CDF command with clear labels for each input.

Normal CDF, lower 0 .68, upper to 70, mean, 0 .64, SD, 0 .2 .7.

Communication is key.

Got it.

Draw, label, shade, calculate, and show your inputs.

Okay.

So we've covered discrete and continuous variables on their own.

What about messing with them, transforming or combining them?

Good question.

Section 6 .2 tackles that.

And the nice thing is, a lot of the rules relate back to what you learned about transforming data way back in chapter two.

Okay.

Refresh my memory.

What happens if you add or subtract a constant, let's say A, to every value of a random variable X?

Adding or subtracting A shifts the entire distribution left or right.

So the measures of center mean median also shift by A.

Makes sense.

What about spread, standard deviation, variance?

That's the key part.

Adding or subtracting a constant has no effect on variability.

The standard deviation, variance, IQR, range, they all stay exactly the same.

The shape of the distribution also doesn't change.

So back to Pete's Jeeps.

His mean collection, C, was $562 .50.

SD was $163 per day.

His present fine.

If he has fixed costs of $100 per day, his profit, V, is C, 100.

Right.

So his mean profit will be medium C, 100, just slides the whole distribution down by $100.

It doesn't make his daily profit any more or less consistent.

Okay.

Adding or subtracting affects center, not spread or shape.

What about multiplying or dividing by a constant, B?

Multiplying or dividing by B affects both center and spread.

The mean, median, standard deviation, and IQR are all multiplied or divided by B.

But the shape stays the same.

If it was skewed right, it's still skewed right, just stretched out or compressed.

Example.

Maybe college tuition.

Sure.

If tuition T is $50 per credit hour X, so T looks 50X.

If the mean number of credits echoes X is 14 .65 and the standard deviation sex is 2 .056.

Then the mean tuition T is 14 .65 equals $732 .50.

Exactly.

And the standard deviation of tuition T is 52 .056, always $102 .80.

Both mean and SD get multiplied by 50.

Now you mentioned something important about variance here.

Yes.

This is crucial.

When you multiply a random variable by B, the standard deviation is multiplied by B, but the variance is multiplied by B square.

BX by VARX is always non -negative anyway.

This B square thing is super important when we combine variables.

Right.

Combining them.

Adding or subtracting two different random variables like X and Y.

Yeah.

Think about Pete's Jeep tours.

Variable X for passengers.

And maybe his sister Erin runs her own tours.

Variable Y for her passengers.

We might care about their total passengers, X plus Y, or the difference, XY.

How do the means work?

Means are easy and always work, regardless of whether X and Y are related or not.

The mean of a sum is the sum of the means.

X plus Y, X plus Y.

And the mean of a difference.

Difference of the means.

XY, XY, XY ways.

If Pete averages 3 .75 passengers and Erin averages 3 .00, their combined average is 3 .75 plus 3 .10 equals plus 3 .10 plus 3 .00 is trickier.

It is.

There's a huge condition you must check before you can combine standard deviations or variances.

The variables must be independent.

Independent.

Meaning, knowing the outcome of one doesn't tell you anything about the outcome of the other.

Exactly.

Like, if Pete and Erin operate in completely separate areas, the number of passengers Pete gets probably has no bearing on Erin's numbers.

Yeah.

We can reasonably assume independence.

Okay, so if they are independent, what's the rule for standard deviation?

Here's the golden rule, sometimes called the Pythagorean Theorem of Statistics.

For independent variables X and Y, variances add always.

Wait, variances add for both sums and differences?

Yes.

That's the surprising part.

So var X plus Y plus var Y and also var XY plus var Y.

Wow.

So variability increases whether you're adding or subtracting independent variables.

That's right.

Because uncertainty or variation from both sources contributes to the overall uncertainty of the sum or the difference.

Think about aiming for a target.

If both your horizontal aim X and vertical aim Y have some wobble, the difference between them XY will also have wobble resulting from both sources.

So critically, standard deviations don't add directly.

Absolutely not.

That's a classic mistake.

You must add the variances first, then take the square root of the sum to find the standard deviation of the combination.

X plus Y, Squartz X2 plus Y2 and Swart X2 plus Y2, assuming independence.

Add variances, then square root.

Got it.

Can you give an example,

distinguishing, combining versus transforming?

Sure.

Think about roulette.

Let X be your winnings from a single $1 bet.

Making two separate $1 bets involves the sum X1 plus X2.

Making one $2 bet involves 2X.

The mean winnings might be the same, but the variability is different.

For X1 plus X2, independent bets, the variance is var X1 plus var X2 equals 2 var X.

For 2X, the variance is 22 var X for 4X.

So the single $2 bet 2X is riskier,

than two separate $1 bets X1 plus X2.

Exactly.

It highlights why you add variances for sums of independent variables, but square the multiplier for transformations.

Okay, this is powerful stuff.

What if the variables we're combining are normal?

Even better news.

If X and Y are independent normal rander variables, then their sum X plus Y and their difference XY are also normally distributed.

Oh, that's huge.

It really is.

Because once you calculate the new mean, X plus Y or XY, and the new standard deviation plus smart way too, you know the result follows a normal distribution.

So you can immediately use normal CDF or Z scores to find probabilities for the sum or difference.

Precisely.

Think about quality control, like fitting lids onto cups.

Maybe cup diameter C is N mean C, SDC, and lid diameter L is N mean L, SDL, and they're independent.

We're interested in the difference D equals LC.

Right.

The mean difference is ND equals LAC.

The variance of the difference is var D equals var L plus var C.

So the standard deviation is D equals NAMASD.

So now you can calculate the probability that the lid fits, which might mean the difference D is positive, lid bigger than cup, but not too big, say 0DA, A along 0, 6 inches.

And you'd use normal CDF with the mean and SD just calculated for D, remembering the drawing, labeling, shading.

You've got it.

It's a common application.

Fantastic.

Okay.

That covers transforming and combining.

Now section 6 .3 gets into two very specific, very important types of distributions.

Right.

Binomial and geometric.

You'll see these a lot on the AP exam because they model many common scenarios involving counting.

Let's start with binomial.

What's the core idea?

Binomial is all about counting the number of successes in a fixed number of independent trials.

Like our bottled water example, counting correct guesses, successes out of 21 students,

fixed trials.

Exactly.

To check if a situation fits the binomial model, we use the mnemonic BNS.

BNS.

Okay.

What's B?

Binary.

Each trial must have only two possible outcomes, which we label success and failure.

Water test.

Correct guess, success.

Incorrect guess, failure.

Independent.

The outcome of one trial can't influence the outcome of another.

One student's guess shouldn't affect the next student's chance if they're truly guessing independently.

Okay.

This is why drawing cards without replacement usually isn't binomial because the probabilities change.

Precisely.

Independence is violated.

N.

Number.

There must be a fixed number of trials, N set in advance.

In the water example, N21 students.

This is why keep drawing cards until you get an ace is not binomial.

The number of trials isn't fixed.

Same probability.

The probability of success, P, must be the same for every single trial.

For the water guessing, if they're just guessing, P equals 13 for everyone.

So BIS, binary, independent, number fixed, same probability.

If all four hold, we have a binomial setting.

Correct.

And the random variable X, which counts the number of successes, is called a binomial random variable.

Its distribution is the binomial distribution, and it's defined just by N and P, often written B and P.

How do we calculate the probability of getting exactly K successes?

P is K.

There's a formula.

P, T, K.

P, P, 1, K.

What's that?

That's the binomial coefficient.

It represents the number of ways you can choose K successes out of N trials.

Your calculator probably has an NCR button for this.

It's calculated as N, K, K.

Okay.

Number of ways times probability of success K times probability of failure.

You got it.

For example, if parents have five children and five, and the chance of typo blood is 0 .25, P of 0 .25, the probability exactly three children have typo blood is?

Five choose three, 0 .25, three, 1 .25, five, three.

Right.

Which is 10, 0 .25, three, seven, five, two.

Comes out to about 0 .0879.

Is there a calculator shortcut?

Yes.

BinomTFNPK calculates PXK, the probability of exactly K successes.

The PDF stands for probability density function, though here it's mass function.

Okay.

BinomTF, what about at least K or at most K successes?

Cumulative probability.

For that, you often use the complement rule or some probabilities.

For PX3, you could find PX4 plus PX5, or more efficiently find PX3 and subtract from one.

Is there a calculator function for cumulative?

BinomCDFNPK.

This calculates PX at K, the probability of getting K successes or fewer.

The CDF is for cumulative distribution function.

Super useful.

So PXC3 would be one PXA.

Exactly.

You'd calculate one BinomCDFNP2.

Mastering the use of BinomDF and BinomCDF, especially with the complement rule, is really important.

And the AP exam tip here is the same as for normal CDF.

Don't just write the function name.

Absolutely.

Show the formula set up with numbers like 5C3, 0 .25, 3 .752, or are clearly labeled the calculator inputs.

BinomDF trials 0 .5, P to jot 0 .25, X value 0 .3.

Yeah.

You have to demonstrate you know what the function is doing and why you're using it.

Got it.

What about the shape, center, and spread of a binomial distribution?

The shape depends on P and N.

If P is close to 0 .5, it's roughly symmetric.

Like flipping a fair coin many times.

Right.

But if P is small, like our P13 water example, the distribution will be skewed to the right.

It's harder to get many successes.

And if P is large, close to 1.

Then it's skewed to the left.

Most trials will be successes.

The larger N is, the less skewed it tends to look, more mound shaped.

Okay.

Center and spread.

Mean and standard deviation.

These have wonderfully simple formulas for binomial.

The mean is just May X of NP equals NP.

So for the water test, N21, P13.

The expected number of successes we'd expect if they were just guessing.

And the standard deviation.

Also straightforward, sex X of score and P1P.

So for the water test, score T, T, NP, 1P.

So for the water test, sick core T, 2113.

Comes out to score it 143, which is about 2 .16.

So we expect seven correct guesses, give or take about 2 .16 guesses, typically.

Okay.

Now let's tie this back to the start.

Mr.

Hogark's class got 13 successes.

Our expected was seven.

SD was 2 .16.

How unlikely was 13?

We need PXE 13.

Using the complement rule, that's 1PXE12.

If you run one binom CDF, 211312.

What do you get?

You get about 0 .00068.

Wow.

Less than 1 % chance.

Exactly.

Such a low probability suggests that the initial assumption, P13, just guessing, is probably wrong.

It provides strong evidence that the students, as a group, could actually tell the difference better than random chance.

That's statistical inference in action.

That's really cool how it connects.

Now you mentioned sampling without replacement violates independence, but there's a workaround.

Yeah, the 10 % condition.

If you're sampling without replacement from a large population, but your sample says N is no more than 10 % of the total population size N, then the change in probabilities from trial to trial is so small that the binomial model is still a very good approximation.

So if you survey 500 teens about debit cards from a population of millions of teens, N500 is way less than 10 % of N, so you can use B500B even though you're not replacing.

Correct.

The 10 % condition lets us treat it as approximately independent.

And there's also a normal approximation for binomial.

Yes.

When N gets large, calculating binomial probabilities directly can be tedious, even with calculators.

If N is large enough, the binomial distribution BNP starts to look very much like a normal distribution.

How large is large enough?

We use the large counts condition.

You need to check that the expected number of successes, NP, is at least 10, and D, the expected number of failures, N1P, is also at least 10.

NPR10 and N1P810.

Right.

If both are true, you can approximate the binomial distribution with a normal distribution that has the same mean, N -way NP, and standard deviation.

Then you can use normal CDF.

Useful for really big sample sizes.

Okay, that covers binomial.

What's the other special distribution?

Geometric.

Geometric.

This one is different.

Instead of counting successes in a fixed number of trials, a geometric variable Y counts the number of trials needed to get the very first success.

So the number of trials isn't fixed anymore?

Nope.

You keep going until that first success happens.

Think about rolling a die until you get a six, or playing a lottery game every day until you finally win.

What are the conditions for a geometric setting?

Very similar to binomial, but without the fixed N.

You still need binary outcomes, success failure, independent trials, and the same probability of success P on each trial.

The variable Y is the number of trials up to and including the first success.

So Y can be one, two, three, with no theoretical upper limit.

Exactly.

You might get lucky on the first try, Y1, or it might take many, many trials.

How do you calculate PYK, the probability that the first success occurs on the K try?

It means you must have K1 failures first, followed by one success.

So the formula is PYK, one past K1P.

Probability of failure K and H1 times probability of success.

For example, a lucky day game where P equals 17 chance of winning each day.

What's the probability your first win happens on the 10th day, Y10?

It would be 67, 10 to 1, 17.

Right.

67, 9, 17, which is about 0 .0357.

Are there calculator functions for geometric?

Yes.

Geometric PDFPK calculates P.

I have the probability that the first success is on the Kth trial, and Geometric PDFP calculates PYK.

The probability the first success happens on or before the Kth trial.

What does the shape of a geometric distribution look like?

Always skewed to the right.

The most likely outcome is succeeding on the first trial, Y1.

The probability decreases as K increases, it gets less and less likely that you have to wait a long time for the first success.

Always right skewed.

Got it.

What about the mean, the expected number of trials until the first success?

Another simple formula.

The mean of a geometric random variable is me equals 1P.

Just 1 divided by the probability of success.

That's it.

So for the lucky day game with P17, the expected number of days you'd have to play until your first win is me 17, 17.

Seven days.

That feels intuitive.

If there's a 1 in 7 chance, you'd expect it to take about 7 tries on average.

Exactly.

It connects nicely.

Wow.

Okay, so that's a pretty thorough tour of random variables, discrete,

continuous, transforming them, combining them, and in these special cases, binomial and geometric.

Yeah, it covers the core concepts from chapter 6.

Understanding these different types and their properties is fundamental for everything that comes later in statistics, especially inference.

We really hope this deep dive has given you some solid footing and maybe some new insights as you continue your AP statistics journey and even beyond that.

Definitely.

And maybe leave you thinking,

how does recognizing these patterns, binomial, geometric, normal, era, help us navigate uncertainty everywhere?

From trying to understand medical trial results to making sense of financial market movements.

It's really about having the tools to quantify and understand chance in a world full of it.

It's pretty powerful stuff.

It absolutely is.

Well, thanks for joining us for this exploration.

Until next time, keep exploring, keep questioning, and keep diving deep.