Statistics
While
not themselves authorities, statistics do resemble authorities in being
frequently quotes as gospel, in adding a transfixing air of knowledgability and
in being misused.
Whenever
making or confronting an appeal of the “statistics
show that...” sort, good thinkers will be prepared to ask a number of
related questions: “Does it reflect
reality?” “Is it complete?” “Is it appropriately precise?” “Are the standards
uniform?” “When were the measurements made?”
Does
it reflect reality eg. quoted the official exchange rate when there is a black
market rate.
Most
statistics achieve meaninglessness by being incomplete. Half the truth, is zero truth. Always demand completeness.
Most
statistics are comparisons. Comparisons
have two or more parts. Demand all
parts.
Always
look for the base from which a claim for a difference is made eg. “30% or more” ==> “30% more than what?”
Statistics
in term of percentages, rates or proportions should usually indicate not only
the base against which a difference is claimed, they should supply absolute
numbers.
Ways
sometimes exist of arriving at statistics which might at first seem impossible
to obtain (via double, or sequence sampling) eg. only 80 of an estimated 800
rapes were reported last year.
Good
statistical reporting sketches methodology.
It take little space and leaves strengths and weaknesses open for all to
see.
Reports
of statistics ought to identify the source.
If
rises or falls in a statistic are to mean much, the standards before the change
and after it must be uniform.
-
watch out for garbage in, garbage out with statistical data gathering and
quality.
System
of significant figures - any expression of a measurement also states the degree
of precision with which the measurement was made.
Like
all classes of statistics, averages have their uses and abuses.
When
we need to emphasise similarity or speak generally, when we want to lump
together or sum up, or when we don’t know details, averages provide valuable
tools.
With
averages you need to look at the standard deviation as well.
Watch
the use of means, modes and medians.
Pictorial
statistics should accomplish visually what written or spoken statistics
accomplish verbally (“a picture is worth a thousand words”)
However
a misleading picture may be worth 10,000 misleading words.
Always
check axes and scales for graphs.
Logarithmic
scales are good for comparing rates which fluctuate from very different
absolute bases eg. compare enrolment fluctuation at a small school with that of
a large school.
Graphic
distortion is really a form of equivocation, usually on relative terms.
Pictographs
are especially prone to manipulation.
Are we to compare relative heights? relative areas? or relative volumes?
No comments:
Post a Comment