How to decide on the number of decimal places to
quote
By MUNGAI KIHANYA
The Sunday Nation
Nairobi,
28 August
2022
The question of
rounding numbers isn’t going away easily. Last week I wrote that I did
not understand why Wafula Chebukati, the Chairman of the Independent
Electoral and Boundaries Commission (IEBC) chose to round off the
percentages to two decimal places. I added that, if I had been
consulted, I would have expressed the figures to just one decimal place.
Due to shortage of
space (this column is just 400 words long), I wasn’t able to explain
why. Rounding off numbers is not done for mere convenience. One must
have good reason for choosing whatever number of decimals.
One such reason is to
ask whether the differences in the numbers can be seen after rounding.
In the case of the election results announced by Mr. Chebukati, it was
clear that even if the percentages were rounded to one decimal place,
the value for each of the four candidates would be different. Thus:
Odinga’s score would 48.8; Ruto’s – 50.5; Wajackoya’s – 0.4 and
Waihiga’s – 0.2 per cent.
However, if we
rounded the figures to the nearest whole number, Wajakoya and Waihiga
would both have zero per cent. This would be misleading – it implies
that they had (nearly) equal number votes. Yet we know that Wajackoya
had almost double what Waihiga got.
These relative scores
are clearly seen when we use one decimal place: Wajackoya – 0.4 and
Waihiga – 0.2. Adding the third decimal place does not give any
additional information. It just opens room for criticism, as we so last
week!
Another reason for
rounding off numbers is to be consistent with margins of errors. In the
case of votes garnered in an election, the values are exact. But in an
opinion poll of a sample of voters, statistical errors are unavoidable.
Obviously, the values
cannot be more accurate than the margin of error, hence they cannot have
more decimal places! Thus, if the estimated error is, say, ±2 per cent,
the results can only be quoted to in whole numbers only. Adding decimal
places implies that the accuracy is greater than the stated margin of
error – “a mathematical absurdity that defies logic”!
Nevertheless,
political opinion pollsters alive to the repercussion of saying that a
certain candidate has zero per cent support! For this reason, it is
understandable to see poll results showing some candidates with less
than one per cent yet the margin of error is greater than one.
|