Chapter 2: Exploring Data with Tables and Graphs

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.

Welcome to the Deep Dive, the show where we take a stack of information, sift through the noise, and really extract the most important nuggets of knowledge, helping you become well informed without all the overwhelm.

And today we're diving into something pretty foundational.

How to unlock the stories hidden inside data.

Think of it as learning the language of numbers so you can spot surprising facts and maybe more importantly, avoid being misled.

Yeah, exactly.

We're drawing our insights from, you know, the bedrock of statistics, specifically how to organize, summarize, and represent data.

And what better way to kick things off than with something relatable?

Traffic.

Oh, don't get me started.

Does it feel like rush hour just keeps expanding like it's most of the day now?

Well, it's not just a feeling.

The U .S.

Census Bureau actually tracked this.

Average commute times went up from twenty six point six to twenty six point nine minutes.

Okay, point three minutes.

That sounds tiny, right?

Almost like nothing.

It seems like it.

But multiply that out over a year of commuting back and forth.

That's an extra two and a half hours the average American is spending stuck in traffic every year.

Wow.

Okay.

So that tiny number actually adds up.

It really does.

And that's exactly why we're doing this deep dive.

If you just look at a raw list, say, 50 commute times in Los Angeles, it's just a jumble.

You can't really see the story.

Right.

So our mission today is to give you the tools, powerful ways to organize, summarize, and crucially visualize data so you can spot those patterns, those shifts, and really get what the numbers mean.

And to help navigate, we focus on five key characteristics of data.

We use the mnemonic CVDOT.

Makes it easier to remember.

CVDOT.

CVDOT.

So C is for center.

That's like your typical value, the middle ground.

Yep.

Then V is for variation.

How spread out are the numbers?

Are they all clustered together or all over the place?

Got it.

D is distribution.

Right.

What's the overall shape when you lay the data out?

Is it lumped in one spot or maybe spread evenly?

Exactly.

O stands for outliers.

Those unusual values way far off from the main group.

They can be really interesting.

And finally, T for time.

How did these characteristics change over, well, over time?

So CVDOT.

That's our framework.

Let's start with the first step in taming that data chaos.

Which is?

It's called a frequency distribution.

Or sometimes just a frequency table.

Basically, you group your data into categories, we call them classes, and then count how many data points land in each category.

Okay.

So back to those 50 LA commute times.

Instead of just a list, we group them.

Like how many commutes were zero to 14 minutes?

How many 15 to 20 minutes?

Exactly.

And suddenly you see patterns.

Table 2 -2 in the book shows this clearly.

Most commutes are, yeah, in that 15 to 29 minute range.

Very few are super long.

You instantly get a sense of the center and the basic distribution.

Right.

It's organized.

So what are the key terms here?

You mentioned classes.

Yeah.

So each class has lower class limits and upper class limits.

The smallest and largest numbers that can belong in that class, like zero and 14 in your example.

Okay.

Then you have class boundaries.

These are the numbers that separate the classes without any gaps.

So if one class ends at 14 and the next starts at 15, the boundary is right in the middle.

14 .5.

It makes sure every value has exactly one home.

Ah.

So no ambiguity.

Precisely.

We also talk about class midpoints.

That's just the middle value of each class.

Add the lower and upper limit, divide by 2.

Easy.

Gives you a single number to represent that whole class.

And class width.

How wide each category is.

Right.

And this is where people sometimes slip up.

If your classes are zero, 14, 15, 29, 30, 44,

the width is 15.

It's the difference between consecutive lower limits,

like 15 minus zero, not 14 minus zero.

Gotcha.

15 minus zero is 15.

30 minus 15 is 15.

Consistent width.

Why is that important?

Well, it keeps the comparison fair across categories.

It helps you get a true sense of the data's variation and how it's distributed.

Okay.

So how do we actually build one of these tables, say with those 50 LA commute times?

It's pretty systematic.

First, decide on the number of classes.

Usually somewhere between 5 and 20 is good.

Let's say we pick 7 for the LA data.

Okay.

7 classes.

Second, calculate the class width.

Take the maximum value minus the minimum value and divide by the number of classes you chose.

So if the longest commute was 90 minutes and the shortest was 5, that's 95, 7.

Which is 85 divided by 7.

About 12 .1.

Right.

Now you usually round that up to a more convenient number.

So 12 .1, we might round that up to 15.

Makes the limits nice and clean.

Okay.

Round up to 15 for the class width.

Then what?

Third, pick your starting point.

The first lower class limit, it should be the minimum value or something convenient just below it.

For commutes, starting at 0 makes a lot of sense.

0.

Nice round number.

Fourth, list the other lower class limits.

Just keep adding the class width.

So 0, then 0 plus 15 is 15, then 15 plus 15 is 30, 45, 60, 75, 90.

Those are the start points for your 7 classes.

Okay.

I see how that builds up.

Fifth, figure out the upper class limits.

Since the second class starts at 15, the first one must end at 14.

Since the third starts at 30, the second ends at 29 and so on.

So you get 0, 14, 15, 29, 30, 44.

Makes sense.

And the last step.

Just tally them up.

Go through your original 50 commute times and mark which class each one falls into.

Then count the tallies for each class.

That's your frequency.

And boom, frequency distribution.

Much better than a raw list.

Way better.

And this isn't just for numbers, by the way.

You could do this for categorical data too.

Qualitative stuff.

Like what?

Think about causes of fatal plane crashes.

Table 2 -3 shows this.

You categorize by cause, pilot, error, mechanical failure, weather, etc.

and count how many crashes fall into each.

What does that show?

It starkly highlights that pilot error is the most frequent cause.

It immediately points to the center of the problem, you could say.

Right.

So variations on this.

Are there other kinds of frequency tables?

Oh, yeah.

A really useful one is the relative frequency distribution.

Instead of raw counts, you show the percentage for each class.

Just divide the class frequency by the total number of data points, then multiply by 100.

So you see what proportion falls into each category.

Table 2 -4 shows this for the LA commutes and is brilliant for comparisons.

Table 2 -5 compares New York and Boise commute times using these percentages.

The difference is just jump out at you.

Boise has a huge percentage of short commutes.

New York is much more spread out, bigger variation, totally different distribution.

I can see how percentages make comparison easier than raw counts, especially if the total number of commutes was different.

For sure.

And there's also the cumulative frequency distribution.

This one adds things up as you go down the classes.

It tells you the total number or percentage of data points that are less than a certain value.

So for LA commutes, it might show, like, how many trips were under 30 minutes total or under 45 minutes.

Precisely.

It shows the accumulation, useful for understanding percentiles and things like that.

So these tables aren't just for organizing.

They help us think critically, too.

Absolutely.

One key thing statisticians often check for is whether the data follows a normal distribution.

Visually, that's the classic bell shape.

Symmetrical.

Symmetrical.

Frequencies start low, increase to a peak in the middle, then decrease back to low again.

If we look back at the LA commute table, table 2 -2, the frequencies go up and then down, but it's definitely not symmetrical.

No.

It's skewed lots of shorter times than a long tail of longer ones.

Right.

So we can immediately say the LA commute data does not fit a normal distribution pattern.

It tells us something important about its distribution and variation.

Okay.

What else can these tables reveal?

You mentioned something about being a data detective.

Ah, yes.

This is cool.

Analyzing the last digits of numbers in your data set.

Look at table 2 -8, pulse rates.

Notice anything odd about the last digit of each rate?

Let's see.

60, 88,

72, 66.

Huh.

They're all even numbers.

0, 8, 2, 6.

Exactly.

If you were measuring a pulse for a full minute, you'd expect a mix of odd and even last digits, right?

And the fact they're all even is a huge clue.

It strongly suggests the person collecting the data probably counted beats for only 30 seconds, and then just doubled the number.

Whoa.

So the table tells you how the data might have been collected, and maybe it's limitations.

Precisely.

Or, say you're looking at survey results for people who report their weight, if you see way too many weights ending in 0 or 5.

People are probably rounding off, not giving their exact weight.

You got it.

It hints at measurement practices.

Another detective trick is looking for gaps in the frequency distribution.

Table 2 -9 shows weights of pennies.

There's a big gap with no pennies in a certain weight range.

Why would there be a gap?

It suggests your data isn't one uniform group.

It's likely coming from two or more different populations mixed together.

Like old pennies and new pennies.

Exactly.

That gap is the clue.

Pennies made before 1983 were mostly copper.

Pennies made after are mostly zinc, much lighter.

The frequency distribution, by showing that gap, reveals this underlying difference, a change related to time that you'd miss just looking at a list.

That's fascinating.

So organizing data really does unlock hidden stories.

It's the essential first step.

Now let's move from tables to actually seeing these patterns.

Let's talk visuals, specifically histograms.

Okay, histograms.

If the frequency table organizes, the histogram paints the picture.

Perfect analogy.

A histogram is a graph.

It uses bars, side by side with no gaps, to show the frequency distribution.

The horizontal axis represents the classes of data, and the vertical axis shows the frequency, how many data points are in each class.

So taller bars mean more data in that range.

Exactly.

Figure 2 -2 shows the histogram for the LA commute times.

You can instantly see that shape we talked about.

Most bars are on the left, shorter commutes, and it tails off to the right.

It gives you a visual gut feeling for the center, the variation, and especially the distribution shape.

And any really short or tall bars that stand alone might flag potential outliers.

And can you do this with percentages too, like the relative frequency table?

Yep, that's a relative frequency histogram, figure 2 -3.

Looks exactly the same shape, but the vertical axis is marked with percentages instead of counts.

Great for comparing datasets, like we said before.

So when we look at a histogram, we're using our CVDOT framework again.

Definitely.

We look at the center, where's the peak.

We look at variation, how spread out are the bars.

We look at the distribution, and what's the overall shape.

We look for outliers, any bars far from the rest.

And if we have histograms from different times, we look for changes over time.

What are the common shapes we might see?

You mentioned the bell shape.

Right, the normal distribution or bell curve, figure 2 -5.

It's symmetrical, peaks in the middle.

Super important in statistics because lots of natural things follow this pattern.

Okay, what else?

You might see a uniform distribution, where all the bars are roughly the same height.

Means all values are about equally likely.

Like rolling a fair die lots of times, each number should come up about the same amount.

Good example.

Then there's skewness.

That means it's not symmetrical.

If the longer tail stretches out to the right, like our LA commutes, figure 2 -2 or 2 -4C, it's skewed right, or positively skewed.

Think annual incomes.

Most people are clustered at lower or middle incomes, but a few high earners pull the tail out to the right.

Okay, skewed right means the tail's on the right.

A mnemonic some people use is to imagine the shape represents your foot.

If it looks like your right foot with the toes pointing right, it's skewed right.

Huh.

Okay, I can picture that.

So skewed left.

Tail stretches to the left, figure 2 -4.

Skewed left or negatively skewed?

Like maybe test scores where most people did well, high scores on the right, but a few did poorly, pulling the tail to the left.

Or lice mans, most people live longer, fewer die young, left foot, skewed left.

Got it.

Histograms make these shakes really obvious.

They do.

For a more rigorous check of normality, statisticians sometimes use something called a normal quantile plot, but the histogram gives you a great first visual assessment.

Okay, so visuals are powerful, which brings us to a really important topic.

Graphs can illuminate, but they can also mislead, right?

Oh, absolutely.

It's crucial to know the difference.

We need to talk about graphs that enlighten and graphs that deceive.

Let's start with the good guys.

What graphs really help us understand data?

Dot plots are fantastic, figure 2 -6.

Super simple.

You just draw a number line and place a dot above the line for each data value.

If values repeat, the dots stack up.

What's good about that?

You see the distribution instantly, you see clusters, gaps, potential outliers,

and critically, you don't lose the original data values.

Sometimes we add jitter, just nudging the dots slightly side to side if they stack up, so you can see every single point clearly.

Nice.

What else?

Stem plots, or stem and leaf plots.

These are clever.

They split each number into a stem, usually all digits except the last, and a leaf, the last digit.

How does that look?

You list the stems vertically and then list the leaves horizontally next to their stem.

It ends up looking like a histogram turned on its side, but again, it keeps all the original data values and it sorts the data for you.

Great for seeing distribution shape quickly with smaller data sets.

Okay, dot plots and stem plots keep the original numbers visible.

What about data over time?

For that, the time series graph is essential, figure 2 -7.

Just plot your data points in chronological order and connect them with lines.

It immediately shows trends, cycles, or sudden changes over time, like the graph showing PC shipments peaking around 2011, then declining vital info for that industry.

And this is where we need to be careful about that fallacy you mentioned.

Ah, yes, the Texas sharpshooter fallacy.

Critical thinking alert.

This is when someone looks at a whole mess of random data, finds one little cluster that happens to look good or fit their story, and then claims it's significant, ignoring everything else.

Like shooting the barn door, then drawing the target.

Exactly, they draw the bullseye after the fact.

You see this when people cherry pick data points from a time series or map to prove a point, conveniently ignoring the data points that don't fit.

It's a misleading way to present distribution or variation.

Good to watch out for.

Okay, what about graphs for categorical data, like the plane crash causes?

Simple bar graphs work well there.

Similar to histograms, but for categories.

Bars represent the frequency of each category,

and crucially, there are gaps between the bars because the categories are distinct.

And the special type, Pareto.

Right, Pareto charts, figure 2 to 8, named after Vilfredo Pareto.

It's a bar graph for categories, but you arrange the bars in descending order frequency, from tallest on the left to shortest on the right.

Why do that?

It instantly draws your eye to the most important categories.

For the plane crashes, pilot error is the tallest bar, way out front.

It helps prioritize tackling the biggest bars, gives you the most impact.

It highlights the center of the categorical frequencies.

Makes sense.

What about the infamous pie chart?

Ah, pie charts.

Figure 2 to 9A, everybody knows them.

Slices of a circle representing percentages or frequencies of categories.

But I sense a but.

Well, yeah.

Many data visualization experts really dislike them.

Edward Tufte, a big name in the field, is famously quoted, never use pie charts because they waste ink on components that are not data and they lack an appropriate scale.

Harsh.

Why so down on them?

His point, and many agree, is that humans are just not very good at accurately comparing angles or areas in slices.

We're much better at comparing lengths, like the bars in a bar graph or Pareto chart.

If you compare the Pareto chart, figure 2 to 8, and the pie chart, figure 2 to 9A, for the same plane crash data, the Pareto makes the relative importance of pilot error much clearer.

Interesting.

So often, better alternatives exist.

Often, yes.

One more enlightening graph is the frequency polygon, figure 2 to 10.

It's similar to a histogram, but instead of bars, you use points plotted above the class midpoints and then connect the points with straight lines.

It emphasizes the continuous change in frequency.

And useful for comparisons.

Especially relative frequency polygons, figure 2 to 11.

Plotting two or more of these on the same axis is a great way to compare distributions.

Like comparing LA and Boise commutes.

The two lines would clearly show LA's frequencies peaking later and being more spread out, confirming the different distributions and variation.

Lots of good tools there.

Now for the dark side.

Graphs that deceive.

What's the biggest culprit?

Probably the non -zero vertical axis,

figure 2 to 12.

This is so common and so misleading.

The vertical axis, the y -axis, doesn't start at zero.

Why is that bad?

Because it exaggerates differences.

Imagine a graph showing nausea rates for two drugs are, say, 20 % and 10%.

A difference?

Sure.

But if the graph maker starts the vertical axis at 10 % instead of 0%, that 10 % difference suddenly takes up the entire visual range of the graph.

The bar for the 20 % drug looks massively taller, maybe 10 or 12 times the size visually, instead of just twice as tall.

It completely distorts the perception of variations.

Sneaky.

So the rule is...

Always check the vertical axis.

Does it start at zero?

If not, be highly skeptical about the visual impression it's giving you.

Okay, check the axis.

What else?

Get to graphs, figure 2 to 13.

These use pictures or icons to represent data quantities.

They look nice, maybe, but are often deceptive, especially if they use 2D or 3D objects to show 1D data.

How does that work?

Remember that NSA phone record example?

Say data collection tripled.

If the artist makes the phone icon three times taller and three times wider to represent this...

Ah!

But area increases much faster than height or width.

Exactly.

If you double the size of a square, the area goes up by four times.

Double the size of a cube, volume goes up eight times.

So a phone that's 3 .5 times taller and wider, like in figure 2 -13, looks maybe 12 times bigger in area.

It massively exaggerates the actual increase, distorting the sense of change over time or the variation.

So be wary of pictures where area or volume is used to show a simple quantity.

Definitely.

Stick to simple bars or lines where length is proportional to the value.

This goes back to Tufts principles for good graph design.

What are those again, briefly?

For small data sets, like 20 or fewer values, just use a table.

It's clear.

Focus on showing the data's true nature.

Don't add distracting chart junk.

Don't distort the data.

Let the graph reveal the truth.

And almost all the ink should be for the actual data, not decoration.

Good principles, honesty and clarity.

It's all about letting the data speak for itself accurately.

Okay, we've covered organizing data, visualizing it and spotting good and bad graphs.

What's next?

Now we get into something really interesting.

Analyzing relationships between two different variables.

Pair data.

This leads us to scatter plots, correlation and regression.

Okay, two variables at once, like height and weight.

Or study, time and test scores.

Exactly.

We ask, is there a correlation?

Meaning,

are the values of one variable somehow associated with the values of the other?

If that association seems to follow a straight line pattern, we call it linear correlation.

And how do we see this?

The first step is a scatter plot.

Just plot the paired x, y data points on a graph.

One variable on the horizontal x -axis, the other on the vertical y -axis.

Simple enough.

Plot the dots.

Yep.

But right here, we need a massive flashing warning sign.

Probably the single most important concept in introductory statistics.

Which is?

Correlation does not imply causality.

Say that again?

Correlation does not imply causality.

Just because two things tend to happen together or move together does not mean one is causing the other.

Give me an example.

Classic one.

Ice cream sales and drowning deaths are correlated.

They both go up in the summer.

Does eating ice cream make you drown?

No.

Obviously not.

It's the heat, the summer weather, that leads to both more swimming and more ice cream eating.

Exactly.

There's a third variable, the lurking variable, warm weather, causing both.

Or think about beer consumption and weight.

There might be a correlation, but you can't just conclude beer causes weight gain from the correlation alone.

Diet, exercise, genetics.

Many other factors are involved.

Mistaking correlation for causation is a huge, huge error in interpreting data distribution and relationships.

Okay.

Crucial point noted.

So scatter plots help us see potential associations.

They do.

Look at figure 214, plotting seal overhead width versus weight.

See how the points form a clear pattern, generally rising from left to right.

Yeah, it looks like wider seals tend to be heavier.

That visual pattern suggests a correlation.

It's why researchers thought they might be able to weigh seals just using photos.

It shows the distribution of the two variables together.

What about no correlation?

Figure 215 shows presidents' heights versus their opponents' heights.

The points are just scattered all over.

No obvious pattern.

Suggest no correlation.

And sometimes scatter plots show clusters, like the pennies.

Figure 216, pennyweight versus year.

You see two distinct blobs of points, the heavier pre -1983 copper pennies and the lighter post -1983 zinc ones.

This shows it's not one relationship, but two different populations defined by time, each with its own distribution.

Ignoring those clusters would be misleading.

So the visual is helpful, but we need something more objective than just looks like a patch.

We do.

That's the linear correlation coefficient, usually written as R.

It's a number calculated from the data that measures the strength and direction of the linear relationship.

Okay, R.

What does it tell us?

R is always between 9 to 1 and plus 1.

If R is close to plus 1, it suggests a strong positive linear correlation.

As one variable goes up, the other tends to go up.

If R is close to my to 1, it suggests a strong negative linear correlation.

As one goes up, the other tends to go up.

What if it's near zero?

If R is close to zero.

It suggests little or no linear correlation.

The points don't really follow a straight line pattern.

How close to one or make one does it need to be to be considered strong?

Ah, good question.

We need objective criteria.

Historically, people compared their calculated R to critical values from a table, like table 211, based on how many data pairs and they had.

If your R was further from zero than the critical value, you'd conclude there was a correlation.

Okay, that sounds a bit fiddly.

Is there a more modern way?

Yes, much more common now is using p -values.

The software that calculates R for you will usually also give you a p -value.

P -value?

What does that mean for correlation?

The p -value here tells you the probability of getting an R -value as strong as far from zero as the one you observed, if there was actually no real linear correlation in the population.

Basically, how likely is it that you saw this pattern just by random chance?

Okay, so a high p -value means?

A high p -value, like say greater than 0 .05, which is a common threshold, means it's quite possible you got that R -value just by random luck, even if there's no real underlying relationship.

So, you'd conclude you don't have sufficient evidence for a linear correlation.

And a low p -value?

A low p -value, typically 0 .05 or less, means it's very unlikely you'd see a correlation that strong just by chance, if there were no real relationship.

So, you conclude you do have sufficient evidence for a linear correlation.

It suggests the pattern you see in your sample reflects a real association.

So small p -value means likely real correlation, large p -value means likely just chance.

That's the core idea.

And sample size plays a big role here.

With only a few data points, like the five shoe print height pairs in table 210, even a moderate R might give you a high p -value, like 0 .294, not enough evidence.

But with more data, like the 40 pairs in figure 218, the same kind of relationship might yield a much stronger R, like 0 .813, and a tiny p -value, like 0 .000, giving you strong evidence for correlation.

More data gives you more power to detect real relationships and understand the true variation.

Okay, so if we do find evidence of a linear correlation, strong R, small p -value,

what can we do with that?

Then we can move to regression.

If there's a linear pattern, we can find the equation of the straight line that best fits through the middle of those points on the scatter plot.

This line is called the regression line.

The line of best fit.

Exactly.

And its purpose is prediction.

Once you have the equation for that line, you can plug in a value for one variable, x, and predict the corresponding value for the other variable, y.

What does the equation look like?

Usually it's written as y -hat plus b1x.

Y -hat means the predicted value of y, b0 is the y -intercept, the predicted value of y when x is 0.

And b1 is the slope, how much y is predicted to change for every one unit increase in x.

So for the shoe print and height data, figure 219, if there's a correlation, you could get an equation like predicted height equals some number plus some other number times shoe print length.

Precisely.

The example gives height equals 80 .9 plus 3 .22 shoe print length.

So if you find a shoe print of a certain length, you can plug it into the equation and get a predicted height.

It's a way to model a relationship and make predictions about the center, or expected value of one variable based on the other.

Wow.

Okay, that's quite a journey we've taken from just a messy pile of numbers.

Right.

We started with raw data, learned how to organize it with frequency distributions, summarize it, and then bring it to life visually with histograms and other graphs.

And we learned to be critical consumers of graphs, spotting the ones that enlighten versus the ones trying to pull a fast one.

Super important skill.

And then we moved into exploring relationships between two variables, seeing patterns with scatter plots, measuring linear strength with correlation R and P values, and finally, making predictions with regression lines.

It really feels like these tools give you a way to systematically understand data using that CVDOT framework center, variation, distribution, outliers, and changes over time.

That's the goal.

Being able to dissect and interpret data, whether it's in tables or graphs,

is, well, it's almost a superpower in today's world, isn't it?

Yeah.

There's just so much information flying around.

It really is.

It's about getting that quick, thorough understanding, spotting the aha moments, and importantly, not getting fooled by bad data or misleading visuals.

So maybe a final thought for everyone listening.

Next time you see a graph, a statistic, a claim based on data, maybe in the news, maybe health info, even just scrolling social media, pause for a second.

Ask yourself, is this really showing me the full picture?

Is it illuminating or is it maybe simplifying things too much or even trying to mislead me?

What's the real story the data is telling?

And critically, how was that data collected and presented?

That's a great takeaway, thinking critically about the numbers and visuals we encounter every day.

We really hope this deep dive helps you explore the world of data with new eyes, maybe more critical eyes.

Yeah, absolutely.

Thank you so much for joining us on this deep dive.

And from the Last Minute Lecture Team, we really appreciate you taking the time to learn with us.