Logistic Regression: A Self-learning Text, Third Edition (Statistics in the Health Sciences)

Disadvantage:

Matching iscostly:

 To find matches

 Information loss due to

discarding controls

Safest strategy:

Matching

Match on strong risk factors expected
to be confounders

MATCHED

(STRATIFIED)

ANALYSIS

STANDARD

STRATIFIED

ANALYSIS

SEE SECTION III

Correct estimate?

YES

Apropriate analysis?

YES

No matching

YES

YES

30–39 40–49

combine

50–59

OR 1 OR 2 OR 3

Validity is not an important reason
for matching (validity: getting the
right answer)

Match to gain efficiency or preci-
sion

III. Matched Analyses
Using Stratification

Strata¼matched sets

The major disadvantage to matching is that it

can be costly, both in terms of the time and

labor required to find appropriate matches

and in terms of information loss due to dis-

carding of available controls not able to satisfy

matching criteria. In fact, if too much informa-

tion is lost from matching, it may be possible to

lose statistical efficiency by matching.

In deciding whether to match or not on a given

factor, the safest strategy is to match only on

strong risk factors expected to cause confound-

ing in the data.

Note that whether one matches or not, it is pos-

sible to obtain an unbiased estimate of the effect,

namely the correct odds ratio estimate. The cor-

rect estimate can be obtained provided an appro-

priate analysis of the data is carried out.

If, for example, we match on age, the appropriate

analysis is amatched analysis,whichisaspecial

kind of stratified analysisto be described shortly.

If, on the other hand, we do not match on age,

an appropriate analysis involves dividing the

data into age strata and doing astandard stra-

tified analysis, which combines the results from

different age strata.

Because a correct estimate can be obtained

whether or not one matches at the design

stage, it follows that validity is not an important

reason for matching. Validity concerns getting

the right answer, which can be obtained by

doing the appropriate stratified analysis.

As mentioned above, the most important sta-

tistical reason for matching is to gain efficiency

or precision in estimating the odds or risk ratio

of interest; that is, matching becomes worth-

while if it leads to a tighter confidence interval

than would be obtained by not matching.

The analysis of matched data can be carried

out using a stratified analysis in which the

strata consist of the collection of matched sets.

394 11. Analysis of Matched Data Using Logistic Regression