Chapter 2: How to Report and Use Uncertainties

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.

Physics is, um, it's usually sold to us as this perfect mathematical machine, right?

Oh, absolutely.

The classic frictionless vacuum.

Yeah, exactly.

The perfectly spherical cows.

Throughout high school, you're taught to imagine this ideal universe.

You run an experiment, plug pristine numbers into a formula, and boom, the universe spits out a single undeniable truth.

Right, right.

Up until you step into an actual real world college laboratory for the first time.

Oh man, the reality check.

Suddenly that pristine machine starts sputtering.

You realize you are wading into a landscape of measurements that is, honestly,

just completely muddy.

It's the absolute definition of diagnostic friction.

Because, well, in a real physics lab,

a measured number without an uncertainty attached to it isn't just incomplete.

I mean, it's essentially meaningless.

Totally meaningless.

So for everyone listening, welcome to this deep dive.

Consider this a personalized tutoring session from the Last Minute Lecture team.

Yeah, we're glad you're here with us.

If you are a student staring down your first real error analysis in a lab, we've got your

We are using Chapter 2 of the Textbook Introduction to Error Analysis to help you master how you actually report and use uncertainties.

And we are moving logically here.

From basic definitions to visual graphs to the ultimate goal, which is...

Propagating those errors when you inevitably have to do the math.

Exactly.

And I promise, we aren't just going to throw complex formulas at you to memorize.

We are going to explore the mechanical reasons why these rules even exist.

We want to turn them from textbook chores into intuitive tools.

Right.

So let's start at the very bottom.

A single measurement.

Yeah, because before we can do any calculating, we have to know how to actually write down our data correctly.

The textbook introduces the standard format.

Your measured value of a variable, let's just call it x equals your best estimate, x best plus or minus your uncertainty, which is delta x.

And that plus or minus symbol is doing a lot of heavy lifting.

Oh, for sure.

Your x best is basically the midpoint of your measurement.

And the delta x is establishing a margin of error.

And by convention, delta x is always a positive number.

Okay, so always positive.

Right.

So when you write this out, you're basically stating that you are reasonably confident the true value of whatever you are measuring lies somewhere between that lower bound and that upper bound.

So it's less like a solid brick wall and more like a weather forecast.

A weather forecast.

How do you mean?

Like if a meteorologist says the hurricane is going to hit somewhere inside this 50 -mile cone, it highly probably will, even if there's a tiny statistical chance it drifts outside of it.

That is a great way to think about it.

What's fascinating here is how universal that concept is.

The text actually compares it to public opinion polls.

Oh, right.

Like when you hear a candidate has 60 percent of the vote with a three percent margin of error.

Exactly.

That margin usually represents a 95 percent confidence limit.

Now, the strict statistical laws governing those confidence intervals, that comes later in your textbook.

Spoiler alert for future chapters.

Yeah, exactly.

But for now,

your delta x is establishing that exact same kind of highly probable boundary for your physical experiment.

Which honestly brings up a massive practical issue for me.

Let's hear it.

Let's say I'm in the lab, right?

I've been calculating my uncertainty for an hour and

out 0 .02385 meters per second squared.

I want to write all of those digits down in my final lab report.

I mean, I worked hard for them.

I know the feeling, but you should resist that urge completely.

Really?

Just throw them out.

Experimental uncertainties should almost always be rounded to one significant figure.

So your 0 .02385 becomes just 0 .02.

Wait, hold on.

If I just chop off the three, the eight, and the five, aren't I technically falsifying my data?

Why throw away that precision?

Because claiming you possess that precision is a logical oxymoron.

Okay, how so?

Well, think about what an uncertainty actually is.

It is a formalized estimate of your own error.

It is absurd to claim, you know, your own ignorance to five decimal places.

Oh, yeah.

When you put it like that, it does sound a bit ridiculous.

Right.

By keeping those extra digits, you are just muddying the waters with phantom data that implies a level of control you simply do not have.

Okay, I see.

I guess the text does offer one small lifeline for my hard -earned digits, though.

The exception to the rule.

Yeah.

If the leading digit of your uncertainty is a one, keeping two significant figures might be less misleading.

And the underlying math completely backs that exception up.

If your calculator gives you an uncertainty of 0 .14 and you round it to 0 .1, you have just artificially reduced your stated margin of error by nearly 30 percent.

Wow.

Yeah, that is a massive proportional drop.

It is.

A leading one is sensitive.

But if the leading digit is a larger number, like a four or an eight, you stick strictly to the one significant figure rule.

Makes sense.

Now, how does that brutal rounding affect my actual best estimate, the X best?

So, the golden rule of error analysis is that the last significant figure in your stated answer must match the decimal position of your uncertainty.

Okay.

Match the decimal position.

Yeah.

Imagine walking into a lab and declaring your measured speed is 6051 .78 plus or minus 30 meters per second.

That makes zero sense.

Exactly.

If my uncertainty is 30, that means the digit in my tens place, the five, could realistically be as low as a two or as high as an eight.

If the tens place is that fuzzy,

the ones place and the decimals are like complete mathematical fiction.

Got on.

Those trailing digits, the one, the seven, and the eight have zero physical significance.

They must be rounded off to match the fuzziness.

So, your correct statement of that measurement is 6050 plus or minus 30 meters per second.

And to save a massive headache, the text suggests using scientific notation here, right?

Oh, absolutely.

It's a lifesaver.

Like, if you are measuring a tiny electrical charge, instead of writing a chaotic string of zeros, you write it as 1 .61 plus or minus 0 .05, all times 10 to the negative 19th coulombs.

It just forces you to align the decimal places visually.

It's much cleaner.

Formatting a single measurement properly sets the stage.

But, you know, science is a team sport.

Usually, yeah.

What happens when two people measure the exact same thing and they get different answers?

Ah, the dreaded discrepancy.

In error analysis, that's defined simply as the numerical difference between two measured best estimates of the same quantity.

Right.

But do we panic every time there's a discrepancy?

I mean, it feels like they're unavoidable.

They are unavoidable.

And we only panic if the discrepancy is significant.

Significant discrepancy.

Let's visualize figure 2 .1 from the chapter.

Student A gets 15 plus or minus 1 ohm.

Student B measures the same resistor and gets 25 plus or minus 2 ohms.

So the discrepancy between their best estimates is 10.

Right.

And 10 is vastly larger than their combined margins of error.

Their individual ranges 14 to 16 for A and 23 to 27 for B, they don't overlap at all.

It's like two compact cars parked on completely opposite ends of a huge parking lot.

That lack of overlap makes the discrepancy significant.

It is a glaring alarm bell telling us that someone, and possibly both of them, made a procedural mistake.

Uh oh.

Back to the drawing board for them.

Exactly.

But on the flip side, what if student C got 16 plus or minus 8 ohms and student D got 26 plus or minus 9 ohms?

The discrepancy between the best estimates is still 10.

Oh, but those margins of error are massive.

The ranges overlap almost entirely.

It's like two giant RVs taking up half the parking lot.

Their boundaries are bound across.

Great analogy.

That discrepancy is insignificant.

Both measurements could technically be describing the same true value within their stated fuzziness.

This gets really interesting when you compare your lab results to a published accepted value.

Right.

Checking your work against the textbook.

Yeah.

Let's say the experiment is measuring the speed of sound in air.

The accepted true value at standard temperature and pressure is 331 m per second.

Student A gets 329 plus or minus 5.

Which is a highly satisfactory result.

The accepted 331 sits comfortably inside student A's margin.

But then student C finishes, student C gets 345 plus or minus 2 m per second.

The discrepancy is 14 m per second.

That is seven times larger than his stated uncertainty.

That's a huge gap.

If I'm student C, I'm terrified my grade is ruined.

Well, before you assume your data is garbage, you have to look for a systematic error.

Not a random error.

No.

A discrepancy that large rarely comes from random fuzziness.

Maybe the timing clock was poorly calibrated.

Or consider the physical environment.

The environment.

Yeah.

Standard temperature is zero degrees Celsius.

But what temperature is a typical college lab usually kept at?

Room temperature.

Which is around 20 degrees Celsius?

Yeah.

The math literally forces you to look at the physical room you're standing in.

Exactly.

At 20 degrees Celsius, the true speed of sound is actually 343 m per second.

If student C had compared his result to the correct accepted value for his specific room conditions, his measurement of 345 would have completely overlapped with the 343.

It's a brilliant example of why critical thinking has to accompany the arithmetic.

You can't just be a calculator.

You really can't.

So that covers comparing to a textbook.

But what if the experiment requires comparing two values you measured yourself?

Like testing a physical law.

Let's take the conservation of momentum.

Theory dictates the initial momentum of two crashing carts.

P should perfectly equal the final momentum, Q.

Right.

But when you measure P and Q, they will both have their own uncertainties.

Eyeballing the ranges to see if they overlap is okay.

But calculating the difference, P minus Q, is far more rigorous.

Because if momentum is genuinely conserved, P minus Q should be exactly zero.

Or, well, at least statistically consistent with zero.

Which creates a new problem for us.

We have to figure out the uncertainty of that calculated difference.

Right.

How do we propagate the error from our two initial measurements into this brand new number we just calculated?

Let's avoid reading the pure algebra out loud and just think about the physical limits.

To find the biggest possible difference between the two.

You'd take your absolute highest possible measurement for momentum P and subtract your absolute lowest possible measurement for momentum Q.

Exactly.

And what happens to the gap?

The gap between them stretches out by the sum of both their errors.

Spot on.

And the same logic applies to finding the lowest probable value.

You take the smallest P and subtract the largest Q.

The result gives us our first major tool from the chapter.

The provisional rule for addition and subtraction.

Okay, lay it on me.

If you are calculating a sum or a difference of two measured quantities, the uncertainty of your result is simply the sum of the absolute uncertainties of your original measurements.

Oh, that's super easy.

So if P had an uncertainty of 0 .03 and Q had an uncertainty of 0 .06, the uncertainty of the difference is simply 0 .03 plus 0 .06.

Which is 0 .09.

Right.

So if my P minus Q calculation comes out to, say, negative 0 .07 plus or minus 0 .09, zero is safely inside that range.

And momentum is conserved.

It is a beautifully simple rule for addition and subtraction.

It really is.

I love that rule, but honestly, if I'm running an experiment with 50 data points, calculating the difference for every single trial and logging it all in a giant table sounds agonizing.

It is entirely impractical, which is why physics relies so heavily on graphical analysis to solve.

The visuals are always better.

It is a visual intuitive way to check relationships between variables across a massive data set.

The textbook uses Hooke's law as the primary example for this.

Right.

Miss springs.

Yeah.

The extension of a spring is proportional to the mass you hang on it.

If you plot the extension on the axis against the mass on the x -axis, the theory demands you get a perfectly straight line passing through the origin.

But again, we are dealing with our muddy reality.

If you just plot your best estimates as tiny dots on a graph, they will scatter.

You have to include error bars.

So instead of a dot, it's a cross.

Exactly.

If your spring extension has an uncertainty of plus or minus 0 .3 centimeters,

you draw a vertical line covering that 0 .6 centimeter range, and then you do the same horizontally for the mass uncertainty.

And this is where the visual power of the graph shines.

Think of a straight line drawn with a ruler as the perfect theoretical universe.

And think of your error bars as our flawed, fuzzy reality.

Okay.

If you can draw a straight line through the origin that passes through, or reasonably close to, all of those error bars,

the theory survives.

Your variables are proportional.

What if the theoretical perfect universe completely misses our fuzzy reality?

The text shows a graph where the best straight line you can draw completely misses the error bars on the heavier masses, like the data clearly curves upward, away from the ruler.

That visual deviation tells a profound physical story.

The law is breaking down.

Oh, because the spring is messed up.

Right.

Perhaps you hung so much mass on the spring that you stretched it beyond its elastic limit, and now it's deforming permanently.

The error bars prove this isn't just random measurement noise.

It is a real physical effect.

That's super cool.

The text also provides a clever trick for relationships that aren't straight lines.

A linearization trick.

Yeah.

Say you were measuring the distance a falling object drops over time.

That's a parabola.

y equals a x squared.

Human eyes are notoriously terrible at judging if scattered data points perfectly fit a curve.

We are awful at it.

We are, however, excellent at judging straight lines.

So the trick is to linearize the data.

How does that work?

Instead of plotting distance against time, you plot distance against time squared.

If line through the origin.

Which allows you to easily check your fit using the ruler and the error bars again.

Exactly.

Okay.

So up to this point, we've been dealing entirely with absolute numbers, plus or minus 0 .3 centimeters,

plus or minus 30 meters per second.

The raw units.

Right.

But context changes everything.

An error of 30 meters per second when measuring the velocity of a distant galaxy is an incredible triumph.

Nobel Prize worthy.

But an error of 30 meters per second when measuring the speed of a bicycle on the street, that is a catastrophe.

Which introduces the critical distinction between absolute uncertainty and fractional uncertainty.

It is the difference between asking how much off are we and asking what is the fundamental quality of this measurement.

The text uses a great distance analogy here.

If you have an absolute uncertainty of one inch in measuring a distance of one mile, that is an incredibly precise, high quality measurement.

But if you have an uncertainty of one inch in measuring a distance of three inches.

That is a completely useless estimate.

Exactly.

The absolute uncertainty is exactly the same in both cases.

One inch.

But the fractional uncertainty tells the real story.

Fractional uncertainty is simply your absolute uncertainty divided by the absolute value of your best estimate.

And since you divide a length by a length, the units cancel out.

It's a dimensionless ratio, usually expressed as a percentage.

Let me think.

So one inch over one mile is a fractional uncertainty of about 0 .0002%.

Tiny.

But one inch over three inches is a fractional uncertainty of 33%.

And this concept ties directly back to our rules for significant figures from earlier.

The chapter lays out a brilliant shortcut.

It's a table.

Yeah.

If your measurement is only good to one significant figure, your fractional uncertainty is somewhere between 10 % and 100%.

Pretty rough.

Very rough.

If you have two significant figures, you're looking at 1 % to 10 % uncertainty.

And three significant figures means a highly precise 0 .1 % to 1 % uncertainty.

That makes fractional uncertainty a really great shorthand for quality.

But why do we actually need it for our math?

Because of multiplication.

Oh.

We know how to add and subtract measurements.

Yeah.

But what if our formula requires us to multiply?

Let's go back to momentum, where momentum equals mass times velocity.

Both mass and velocity have their own uncertainties.

How do we find the uncertainty of the final multiplied momentum?

Well, we can't just add the absolute uncertainties together, right?

I mean, adding an error in kilograms to an error in meters per second is like adding apples to spark plugs.

Exactly.

The units don't match.

This is where fractional uncertainties become the stars of the show.

Instead of reading out the brutal algebra from the textbook,

let's visualize the mechanism.

Okay, I'm picturing it.

Imagine scaling the base values.

When you multiply the mass and the velocity together, their relative fuzziness stacks.

You get the tiny percentage fuzziness of the mass, the tiny percentage fuzziness of the velocity.

And then there's a leftover mathematical piece where the two tiny fuzzinesses multiply together.

Ah, so it's the fuzziness of the fuzziness.

Like, if my mass has a 2 % error and my velocity has a 3 % error, that leftover piece is 2 % of 3%, which is 0 .0006.

It's microscopically small.

And because that final stacked term is so incredibly tiny compared to the main errors, we can safely neglect it.

We just drop it from the equation entirely.

Wow, just throw it out.

That leaves us with the second major tool, the provisional rule for multiplication and division.

Yep.

When you multiply or divide two measured quantities, the fractional uncertainty of the result is the sum of the fractional uncertainties of the original quantities.

You just add the percentages.

Just add them up.

If mass has a 2 % uncertainty and velocity has a 3 % uncertainty, your calculated momentum has a 5 % uncertainty.

And then to find your final absolute uncertainty, you just calculate 5 % of whatever your final momentum value was.

It creates a perfectly symmetrical toolkit for any student doing error propagation.

Use some absolute errors for addition and subtraction and use some fractional errors for multiplication and division.

That is actually really elegant.

Let's pull all of this together before you, the listener, step into the lab.

Let's do a quick recap.

First, always use the standard format, your best estimate plus or minus your uncertainty.

Second, round that experimental uncertainty to one significant figure and force your best estimates final digit to perfectly match that decimal position.

No phantom data.

Third, rely on graphical analysis.

Plotting your data with proper error bars will instantly show you whether discrepancies in your data are just acceptable measurement fuzziness or if a physical law is fundamentally breaking down in front of you.

And finally, keep those two provisional rules for propagating errors close at hand.

Oh, and remember to look at the physical room before you assume your data is garbage.

Always check the room temperature.

But, you know, I do want to leave you with one final thought to mull over regarding that multiplication rule we just talked about.

Oh, a parting thought.

Lay it on us.

We dropped that final mathematical term, the fuzziness of the fuzziness, because we assumed it was microscopically small.

We assumed you were working with high quality measurements of 2 % or 3 % error.

Right, the point 0006 just vanishes into the statistical static.

But what if your lab goes horribly wrong?

What if your measurements are so crude that your fractional uncertainty is huge, say 50 % for both variables?

Oh, man.

The math radically changes.

50 % of 50 % is 25%.

That final term isn't microscopic anymore.

It's a massive quarter of your entire measurement that you're just throwing away.

Wait, really?

Really.

Yeah.

Think about how that shatters our neat little equation and think about why the textbook explicitly calls these only provisional rules.

Oh, wow.

If the data is bad enough, the shortcut literally breaks.

That's an incredible note to end on.

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

On behalf of the Last Minute Lecture team, we wish you the absolute best of luck in your introductory physics laboratory.

You've got this.