Write the exact model
A=10000×1.17=19487.171 One exact evaluation of the power gives 19487.171, which rounds to 19487.
Look at the first rounding
10000×1.1=11000 (exact) Year 1 happens to be exact, so no error yet.
Follow the exact chain
11000→12100→13310→14641→16105.1→17715.61→19487.171 The true running values are not whole numbers from year 5 onwards.
See where rounding first bites
16105.1→16105 (0.1 discarded) Rounding to a whole person at year 5 throws away 0.1 of a person.
Follow the rounded chain forwards
16105×1.1=17715.5→17716 The rounded year-5 figure is now the INPUT to year 6, and 17715.5 rounds up to 17716 — the error has changed sign and grown.
Compare the two answers
rounded chain: 17716×1.1=19487.6→19488=19487 The year-by-year rounded chain finishes at 19488, one person above the exact answer of 19487.
Identify the cause
each rounding is fed back into the next multiplication The error is not just displayed — it becomes the input to the next step and is multiplied again.
State the correct method
compute P×rn once, then round Evaluate the whole power at full precision and round only the final answer.
Check the correct method
10000×1.17=19487.171→19487 A single exact calculation avoids all intermediate rounding.
Note when it matters most
large n and tight targets The more steps there are, the more roundings compound — and near a boundary the drift can flip a "first year above" answer.
Rule out a different explanation
1.17 is not the problem The formula itself is right; only the arithmetic method is at fault.
Rule out the multiplier being wrong
r=1.1 is correct for 10% growth A 10% rise really is a multiplier of 1.1.
Rule out the exponent being wrong
n=7 is correct for 7 years Seven years means seven applications of the multiplier.
Summarise the principle
round ONCE, at the end Every displayed intermediate value should be a rounding of the true running value, never a re-multiplication of a rounded one.
State the explanation
rounding each year feeds the error forward and it compounds The year-by-year rounding introduces an error that is then multiplied again in every later year.