Minerva Project Learning Resources Updated June, 2023

Estimation
Introduction
The world is full of numbers that we simply don’t know. How many apples were sold in
total at all the supermarkets in London in October? How many butterflies are there in
Mexico? You might be able to look up some of these numbers, but it’s unlikely that
someone stood in the fruit section of all London grocery stores and counted all the apples
that left the stores for a month, or that all the supermarket chains bothered to count the
number of apples that left their stores and then all come together to aggregate those
numbers into a single statistic for October. What are the chances that someone has
roamed across Mexico counting all the butterflies in the country and given us an easy,
researchable answer? Even if they did, it is difficult to believe that that number would be
accurate. So how can we find the answers to these questions? The short answer is
estimation.

Estimation
Not a lot of people in the world are interested in the answers to these questions, which
means that there is little hope that someone has run a well-researched study on the apple
consumption of Londoners. But what if you’re trying to optimize the number of apples
imported into London so that you can reduce import costs and produce waste. What if
you’re studying butterfly migration patterns in North America, and want to know if your
assumptions are reasonable? Estimation is an excellent tool for creating ballpark
numbers to answer your questions; they can satisfy a sudden curiosity, but they can also
provide important foundations for larger questions and studies.1

Fermi estimates, which are back-of-the-envelope calculations that can be done with a
pencil, paper scrap (and nowadays, usually Google), are named after Enrico Fermi, a 20th
century Italian physicist who significantly furthered nuclear reactor research, and had a
strategy for estimating large numbers with little data. Fermi estimates can be used to
calculate results for all kinds of questions, including the ones above.2

Fermi estimates rely on identifying a set of assumptions that, if you have the numbers,
allow you to perform simple arithmetic to calculate the answer. The classic example of a

1
Helfand, D. (n.d.). Chapter 2: Discoveries on the back of an envelope. In Frontiers of Sciences: Habits of a Scientific Mind.
2
Muehlhauser, L. (Apr 11, 2013). Fermi estimates.

©2023 Minerva Project Inc. Proprietary Information. All Rights Reserved. minervaproject.com
 Minerva Project Learning Resources Updated June, 2023 2

Fermi estimate is a similarly obscure question: how many piano tuners are there in
Chicago?3

An example solution to this problem, through Fermi estimation, would give 6

assumptions:

1. The population of Chicago is about 5,000,000.

2. On average, there are two people per Chicago household.
3. Roughly one household in twenty has a piano that is tuned regularly.
4. Pianos that are tuned regularly are tuned on average about once per year.
5. It takes a piano tuner about two hours to tune a piano, including travel time.
6. Each piano tuner works eight hours a day, five days a week, and 50 weeks per
year.

Once again, these numbers are estimates. Some of them, such as the population of
Chicago, are easy to look up. If you can look up a number, do so! It will only improve your
estimation. However, don’t be afraid to make an educated guess as well.

With these numbers, the next step involves some basic math:

The number of piano tunings in a year in Chicago

𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛
= ( 𝑃𝑒𝑜𝑝𝑙𝑒 𝑝𝑒𝑟 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 ) × (𝑃𝑖𝑎𝑛𝑜𝑠 𝑝𝑒𝑟 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑) × (𝑃𝑖𝑎𝑛𝑜 𝑡𝑢𝑛𝑖𝑛𝑔𝑠 𝑝𝑒𝑟 𝑦𝑒𝑎𝑟)

or:

5,000,000 1
( 2
) × ( 20 ) × 1 = 125, 000 𝑝𝑖𝑎𝑛𝑜 𝑡𝑢𝑛𝑖𝑛𝑔𝑠 𝑝𝑒𝑟 𝑦𝑒𝑎𝑟 𝑖𝑛 𝐶ℎ𝑖𝑐𝑎𝑔𝑜

If the average piano tuner works:

(50 𝑤𝑒𝑒𝑘𝑠 𝑝𝑒𝑟 𝑦𝑒𝑎𝑟) × (5 𝑑𝑎𝑦𝑠 𝑝𝑒𝑟 𝑤𝑒𝑒𝑘) × (8 ℎ𝑜𝑢𝑟𝑠 𝑝𝑒𝑟 𝑑𝑎𝑦) × (2 ℎ𝑜𝑢𝑟𝑠 𝑝𝑒𝑟 𝑝𝑖𝑎𝑛𝑜 𝑡𝑢𝑛𝑖𝑛𝑔) = 1000

Then there are 1000 piano tunings per year per piano tuner.

Finally:

(125, 000 𝑝𝑖𝑎𝑛𝑜 𝑡𝑢𝑛𝑖𝑛𝑔𝑠 𝑝𝑒𝑟 𝑦𝑒𝑎𝑟 𝑖𝑛 𝐶ℎ𝑖𝑐𝑎𝑔𝑜) ÷ (1, 000 𝑝𝑖𝑎𝑛𝑜 𝑡𝑢𝑛𝑖𝑛𝑔𝑠 𝑝𝑒𝑟 𝑦𝑒𝑎𝑟 𝑝𝑒𝑟 𝑝𝑖𝑎𝑛𝑜 𝑡𝑢𝑛𝑒𝑟)

= 125 𝑝𝑖𝑎𝑛𝑜 𝑡𝑢𝑛𝑒𝑟𝑠 𝑖𝑛 𝐶ℎ𝑖𝑐𝑎𝑔𝑜

We cannot be sure that this is completely accurate, but it certainly gives us something to
start with, and is more accurate than just a guess. What if we wanted to know the answer
to the London apple problem? This could be as simple as starting with some basic
numbers: how many supermarkets are there in London? How many people visit these
supermarkets each day? How many of them buy apples? If you find the number of apples

3
The Classic Fermi Problem. (n.d.) Portland State University.

©2023 Minerva Project Inc. Proprietary Information. All Rights Reserved.

 Minerva Project Learning Resources Updated June, 2023 3

sold per supermarket per day and multiply that by the number of supermarkets in London
and the number of days in October, you’ll have your answer.

Bounded Estimation
In some cases, however, it really is difficult to come up with an estimate. In this case,
bounding your estimate can significantly help your accuracy. Bounding can be
approached by finding the reasonable lowest and highest numbers for the quantity you
are trying to estimate – if you’re trying to estimate the weight of an apple, you know it is
heavier than a grape and lighter than a watermelon. However, the tighter you can make
the bounds the better, so you may want to refine this further – you know your apple is
heavier than a kiwi fruit and lighter than a grapefruit. For some reason, you know that the
weight of a kiwi is about 80 grams, and a grapefruit is about 225 grams. To calculate an
estimate using bounds, you calculate the geometric mean of your two bounds.4

𝐺𝑒𝑜𝑚𝑒𝑡𝑟𝑖𝑐 𝑚𝑒𝑎𝑛 = (𝑙𝑜𝑤𝑒𝑟 𝑏𝑜𝑢𝑛𝑑) 𝑥 (𝑢𝑝𝑝𝑒𝑟 𝑏𝑜𝑢𝑛𝑑), which is:

80 × 225 = 18, 000

then,

18, 000 = 134.

Our estimate for the weight of the apple is 134 grams.

Now in real life, you could just look up the weight of an apple and know that it’s about
100 grams on average (so fairly close to the estimate!) but this illustrates the theory of
bounded estimates. How could we make this more accurate? Try to find bounds that are
equidistant from the number that you’re looking for. We know that a kiwi is much closer in
size to an apple than a grapefruit; try to take a quantity that has a similar absolute size
difference. If we retry the calculation with an orange, which weighs about 140 grams, then
we take:

(80 𝑥 140) = 106

106 grams. Pretty close!

Dimensional Analysis
Dimensional analysis is a method of unit conversion, and a skill that can significantly
help you with the estimation of quantities with different units of measurement.
Dimensional analysis works with any system of measurement, including the Imperial
System of measurement, the International System (SI), and for conversions between.5

4
Estimates and Fermi Calculations. (n.d.). University of Central Florida.
5
Dimensional Analysis. (n.d.). Chemistry Talk.

©2023 Minerva Project Inc. Proprietary Information. All Rights Reserved.

 Minerva Project Learning Resources Updated June, 2023 4

Much like with Fermi estimation, you will often need to rely on the internet to help you find
the relationships between these units (e.g. that 1 kilogram = 2.2 pounds, or 1 mile = 1.6
kilometers).

When you have identified these relationships, you can make basic unit conversions. For
example, if you are trying to figure out how many pounds a 50 kilogram pumpkin weighs,
you can use dimensional analysis. Your equation would be set up like this:

50 𝑙𝑏𝑠 1 𝑘𝑔
1
× 2.2 𝑙𝑏𝑠
= 22. 7 𝑘𝑔

How does this work? We want to set up all of our units as fractions, so that we can cancel
out repeat units (in this case pounds) until we can narrow it down to only our target unit
(kilograms). 50 pounds is our given number that we are trying to convert to kilograms; to
make it a fraction, we simply make the denominator 1. Then, we multiply it by the
conversion of pounds to kilograms – 1 kilogram = 2.2 pounds. In order to cancel units in a
fraction, you need to be able to cancel a unit from both the numerator and the
denominator, so we set pounds on the bottom and kilograms on the top.6 Then, we can
cancel out pounds so that our equation looks like this:

50 1 𝑘𝑔
1
× 2.2
= 22. 7 𝑘𝑔

Now that we have eliminated the old unit, we can solve the equation by multiplying across
(50 x 1) and dividing the remaining fraction (50/2.2). This gives us the weight of the
pumpkin in kilograms, 22.7.

Dimensional analysis also works when you use more than two units, and/or when you
want to retain more than one unit at the end (e.g. meters per second). For example, if
sound travels at 343 meters per second, how fast does sound travel in miles per hour?
Our equation then looks like this:

343 𝑚 1 𝑚𝑖𝑙𝑒 60 𝑠 60 𝑚𝑖𝑛

1𝑠
× 1609 𝑚
× 1 𝑚𝑖𝑛
× 1 ℎ𝑜𝑢𝑟
=

We keep adding new units with known conversions until both of our target units are part
of the equation (miles and hours). Now let’s cancel the repeat units, which are meters,
seconds, and minutes:

343 1 𝑚𝑖𝑙𝑒 60 60
1
× 1609
× 1
× 1 ℎ𝑜𝑢𝑟
=

Now we can multiple across, and divide:

1,234,800 𝑚𝑖𝑙𝑒𝑠
1609 ℎ𝑜𝑢𝑟𝑠
= 767 𝑚𝑖𝑙𝑒𝑠/ℎ𝑜𝑢𝑟

6
Khan, S. (n.d.). Intro to dimensional analysis. Khan Academy.

©2023 Minerva Project Inc. Proprietary Information. All Rights Reserved.

 Minerva Project Learning Resources Updated June, 2023 5

What if you don’t know the unit conversions? Of course, your dimensional analysis will
always be most accurate with exact conversions, but estimation is not out of the
question. You can use these methods separately or together to arrive at an approximate
answer to many questions with units and numbers.

References
Dimensional Analysis. (n.d.). Chemistry Talk.

https://chemistrytalk.org/dimensional-analysis/

Estimates and Fermi Calculations. (n.d.). University of Central Florida.

https://pressbooks.online.ucf.edu/phy2048tjb/chapter/1-5-estimates-and-fermi-calculation
s/

Helfand, D. (n.d.). Chapter 2: Discoveries on the back of an envelope. In Frontiers of Sciences:

Habits of a Scientific Mind.

http://ccnmtl.columbia.edu/projects/mmt/frontiers/web/index2.html

Khan, S. (n.d.). Intro to dimensional analysis. Khan Academy.

https://www.khanacademy.org/math/algebra/x2f8bb11595b61c86:working-units/x2f8bb11
595b61c86:rate-conversion/v/dimensional-analysis-units-algebraically

Muehlhauser, L. (Apr 11, 2013). Fermi estimates.

http://lesswrong.com/lw/h5e/fermi_estimates/

The Classic Fermi Problem. (n.d.) Portland State University.

https://web.pdx.edu/~pmoeck/pdf/The%20classic%20Fermi%20problem.pdf

©2023 Minerva Project Inc. Proprietary Information. All Rights Reserved.