# Thinking in estimates: units, scaling and educated guessing: When an estimate is good enough to decide

Source page: https://www.materialinsights.org/learn/engineering/thinking-in-estimates/p5/

[All courses](https://www.materialinsights.org/learn/) / [Understanding Engineering](https://www.materialinsights.org/learn/engineering/) / [Module 2: Thinking in estimates: units, scaling and educated guessing](https://www.materialinsights.org/learn/engineering/thinking-in-estimates/) Part 5 of 5

## When an estimate is good enough to decide

An estimate has done its job when the decision no longer depends on the part you do not know. This part shows how errors combine, how to compare an estimate with a real value and read the difference, and ends with a full decision estimate: whether it is cheaper to run an electric scooter or a petrol one.

### 5.1 Errors multiply, so ask how far off you can be before the decision flips

Suppose an estimate multiplies four numbers and each could be wrong by 30 per cent. If every error pushes the same way, the answer is wrong by 1.3 x 1.3 x 1.3 x 1.3, nearly a factor of three. But errors in different inputs rarely line up. When they are independent, the typical combined error grows only as the square root of the number of factors, here twice the single error, about 60 per cent. A four-step estimate with honest inputs is therefore usually good to within a factor of two, and rarely worse than three.

Powers change the arithmetic. A quantity that depends on the square of a length doubles the relative error of that length: a 10 per cent mistake in a radius becomes a 21 per cent mistake in the area it covers. A square root halves it: a 40 per cent error in the depth of a tank becomes a 20 per cent error in the speed of the jet. Knowing which inputs enter with high powers tells you where care is worth spending.

The real question is not how accurate the answer is but whether the decision changes. Put your estimate beside the threshold that would flip your choice. If a plant needs 3 MW and the grid connection offers 10, being wrong by a factor of two changes nothing. If the plant needs 8 MW and the connection offers 10, a 30 per cent error matters, and you should improve the input that contributes the most uncertainty, not the one that is easiest to refine.

Write a range as well as a central value: 'about 4 kWh a day, between 3 and 5.5.' If the whole range sits on one side of the threshold, stop. If it straddles it, you have found what extra information is worth paying for.

Data How input errors reach the answer

| Case | Error in input | Error in result |
|---|---|---|
| Four factors multiplied, errors all in one direction | 30 per cent each | up to a factor of 2.9 |
| Four independent factors, typical | 30 per cent each | about 60 per cent |
| Result depends on length squared | 10 per cent in length | about 21 per cent |
| Result depends on square root of length | 40 per cent in length | about 20 per cent |
| Result depends on one over length | 20 per cent in length | about 17 to 25 per cent |

Calculated from 1.3 to the power 4 = 2.86; 30 per cent x square root of 4 = 60; 1.1 squared = 1.21; 1.4 to the power 0.5 = 1.18; 1/1.2 = 0.83 and 1/0.8 = 1.25.
 Source: Mahajan, Street-Fighting Mathematics (MIT Press, 2010), ch. 5, 'Taking out the big part' (fractional changes); table compiled for this course  Worked example

#### Whether and how much are different questions

A factor-of-two estimate usually answers 'whether' questions: whether to pursue a project, whether a design is plausible, whether a quoted price is in the right region. It does not answer 'how much' questions that involve ordering material or sizing a safety-critical part. Sizing a lift rope, as module 1 showed, needs real strengths, real loads and a code-set margin. Use the estimate to decide that you need the detailed work, then do the work.

- Mahajan, Street-Fighting Mathematics (MIT Press, 2010) · ch. 5, 'Taking out the big part' (small fractional changes and powers)
- Kosky et al., Exploring Engineering (Academic Press, 2013) · ch. 2, 'Elements of Engineering Analysis' (significant figures and uncertainty)

### 5.2 Check every estimate against a real value, and read the miss

An estimate that has never met a real number is still a guess. Every worked case in this module was checked against a published figure, and the habit is worth copying. Look for a number from a regulator, a standards body, national statistics or a maker's datasheet, and note the year. Then make sure it measures the same thing. The LPG total included restaurants. The motor in module 1 had a shaft rating and an electrical input, which differ by the efficiency. A check against the wrong definition can mislead as much as no check.

The check should also be independent of your assumptions. Two different routes to the same quantity are worth more than a hundred repetitions of one.

Next, read the miss. Within about 30 per cent, accept the estimate. A factor of two usually means a missing efficiency or a double count. A factor of ten points to a slip in units, often kilowatts against kilowatt-hours. A factor of a thousand almost always means a wrong prefix. Never change an input just to force a match.

Sometimes the miss is the finding. The 200-litre geyser estimate was nearly twice the label limit, and the explanation, thicker insulation, is the kind of lesson from practice that module 1 described.

Data Reading the size of a miss

| Estimate against real value | Likely cause | What to do |
|---|---|---|
| Within about 30 per cent | Ordinary uncertainty in the inputs | Accept it and say how well it was checked |
| Off by about 2 | Missed efficiency, double count, rough assumption, different definition | Check what the real value includes; look for a missing factor of 0.5 to 0.8 |
| Off by about 10 | Unit slip: kW and kWh, lakh and million, litres and cubic metres | Redo the chain with every unit written beside its number |
| Off by about 1,000 | Wrong prefix: kilo, mega, milli | Check each prefix against the table of powers of ten |
| Off in the 'wrong' direction repeatedly | A real physical effect you left out | Treat the miss as a finding and look it up |

 Source: Kosky et al., Exploring Engineering (Academic Press, 2013), ch. 2, 'Elements of Engineering Analysis' (checking answers and units); table compiled for this course  Common misreading

#### 'If it matches, the method is proved'

A match can be luck. The LPG estimate landed within 5 per cent partly because errors in the pieces cancelled; with the exact customer count it was 15 per cent high. A close match supports the estimate but does not prove each input. A good habit is to state, next to the match, the one input you trust least.

- Kosky et al., Exploring Engineering (Academic Press, 2013) · ch. 2, 'Elements of Engineering Analysis' (the need-know-how-solve method and checking)
- Mahajan, Street-Fighting Mathematics (MIT Press, 2010) · Preface and ch. 2, 'Easy cases'

### 5.3 A decision estimate: what does it cost to run a scooter?

To finish, put the tools together on a household decision: which costs less to run each kilometre, a petrol scooter or an electric one?

Start with the big part. At a steady 40 km/h, the rider and scooter, about 200 kg together, meet two resistances: the rolling of the tyres, roughly 30 N, and the push of air, roughly 45 N. These are assumed round figures that module 3 will let you derive. The total is 75 N. Energy equals force times distance, so each kilometre takes 75 N x 1,000 m = 75 kJ, which is 21 Wh. The motor and controller are about 80 per cent efficient, so the battery gives out 26 Wh. Charging loses about a tenth, so 29 Wh comes from the wall, which we round to 30 Wh/km.

At an assumed ₹8 per unit, this is ₹0.24 a kilometre. For petrol, assume a round 45 km per litre and a price of ₹105 a litre, inside the ₹102 to ₹112 range in the four big metros in November 2026. That is ₹2.3 a kilometre, ten times as much.

Check the electric figure against a maker's data. Ather sells its Rizta with installed batteries of 2.7 or 2.9 kWh, certified for 123 km, or 3.5 kWh, certified for 161 km; its own on-road estimates are 100 and 125 km. The certified numbers give about 22 to 24 Wh/km at the battery and the on-road ones about 27 to 29, if the shorter on-road figure goes with the smaller pack. Our 26 Wh at the battery sits between them.

Now the decision. Even at ₹12 a unit and 40 Wh/km, the electric scooter would cost ₹0.48 a kilometre, five times less, so running cost does not decide between them. Purchase price, battery life and access to charging are, and each needs its own estimate. See 'Motorcycle and scooter' and 'Lithium-ion battery' on the site.

Data Worked estimate: running cost per kilometre, electric and petrol scooter

| Step | Working | Result |
|---|---|---|
| Resistance at 40 km/h | Rolling about 30 N plus air about 45 N, for a 200 kg rider and scooter (assumed; the maker gives a kerb weight of 125 kg) | about 75 N |
| Energy per km at the wheel | 75 N x 1,000 m = 75 kJ; 75,000 / 3,600 | about 21 Wh |
| At the battery | Drive efficiency about 80 per cent: 21 / 0.8 | about 26 Wh |
| From the wall | Charging efficiency about 90 per cent: 26 / 0.9 | about 29, round to 30 Wh/km |
| Electric cost per km | 0.030 kWh x ₹8 per unit (assumed) | about ₹0.24 |
| Petrol cost per km | ₹105 per litre / 45 km per litre (assumed) | about ₹2.3 |
| Ratio | 2.3 / 0.24 | about 10 times |
| Check against a real value | Maker's data: 2.7 or 2.9 kWh for 123 km and 3.5 kWh for 161 km certified (about 22 to 24 Wh/km); 100 and 125 km on the road (about 27 to 29 Wh/km) | Estimate at the battery, 26 Wh/km, falls between the two |

Steady-speed estimate: stops and starts, hills and wind raise real energy use. The 45 km per litre and ₹8 per unit are assumed round figures. The pairing of battery with certified range is from a July 2026 report; the pairing of the on-road figures with the packs is assumed. Ranges are divided by installed, not usable, capacity.
 Source: Ather Energy, Rizta specifications page, installed battery 2.7, 2.9 or 3.5 kWh; IDC range 123 and 161 km; TrueRange 100 and 125 km; kerb weight 125 kg, https://www.atherenergy.com/rizta/specifications  In the data

#### Why the answer is robust

The two costs differ by a factor of ten. Our estimate of the energy per kilometre could be wrong by 50 per cent, electricity prices by 50 per cent, and petrol mileage by 30 per cent, and the ordering would still hold. A decision that survives errors larger than any plausible estimate error has been settled by the estimate. If the gap had been 20 per cent, the same inputs would settle nothing, and the right step would have been to find better data.

- Ather Energy, Rizta specifications page · installed battery 2.7, 2.9 or 3.5 kWh; IDC range 123 and 161 km; TrueRange 100 and 125 km; kerb weight 125 kg, https://www.atherenergy.com/rizta/specifications
- Autocar India, 'Ather scooters get new battery packs, LFP cells on Rizta' (16 July 2026) · Rizta 2.7 kWh (LFP) or 2.9 kWh (NMC) for 123 km IDC; 3.5 kWh for 161 km IDC, https://www.autocarindia.com/bike-news/ather-scooters-get-new-battery-packs-lfp-cells-on-rizta-440234
- HDFC Sky, 'Petrol, diesel prices today, November 11, 2026' (from Goodreturns) · petrol Rs 102 to Rs 112 a litre in Delhi, Mumbai, Bengaluru and Chennai, https://hdfcsky.com/news/petrol-diesel-prices-today-november-11-2026-check-latest-rates-in-delhi-mumbai-kolkata-chennai-bengaluru-hyderabad
- Mahajan, Street-Fighting Mathematics (MIT Press, 2010) · ch. 5, 'Taking out the big part'

### See it in the table
[Li Lithium](https://www.materialinsights.org/commodities/lithium/)[El Electricity](https://www.materialinsights.org/energy/electricity/)

### Check your understanding

Four independent factors in an estimate are each uncertain by 30 per cent. What is the typical error in the product?
 Independent errors add roughly as a square root of the number of factors, so four factors at 30 per cent give about 2 x 30 = 60 per cent. All pulling the same way, the error could reach a factor of 2.9.

An estimate of a plant's power need is a factor of 1,000 away from a published value. The most probable explanation is:
 Errors of exactly a thousand almost always come from a prefix, as ordinary assumptions rarely produce errors that large.

An electric scooter costs ₹0.24 per km to run and a petrol one ₹2.3. Which statement best describes the value of the estimate?
 A tenfold gap survives errors of 50 per cent in every input, so the running cost ordering is clear. The purchase decision depends on other factors that would need their own estimates.

[← Part 4: Scaling: why size changes everything](https://www.materialinsights.org/learn/engineering/thinking-in-estimates/p4/)[Mark read · Wrap-up →](https://www.materialinsights.org/learn/engineering/thinking-in-estimates/review/)

## Other sections

- [Module overview](https://www.materialinsights.org/learn/engineering/thinking-in-estimates/index.md)
- [Why engineers estimate before they calculate](https://www.materialinsights.org/learn/engineering/thinking-in-estimates/p1/index.md)
- [Dimensions and easy cases: checking formulas and finding them](https://www.materialinsights.org/learn/engineering/thinking-in-estimates/p2/index.md)
- [The big part first, then rounding, lumping and pictures](https://www.materialinsights.org/learn/engineering/thinking-in-estimates/p3/index.md)
- [Scaling: why size changes everything](https://www.materialinsights.org/learn/engineering/thinking-in-estimates/p4/index.md)
- [Key terms, quiz and sources](https://www.materialinsights.org/learn/engineering/thinking-in-estimates/review/index.md)
