Introduction to Statistics
This page of the CIRCLIST website brings together in one place a number of survey results and statistical analyses that previously were spread around the site. Before delving into the detail, you may care to read the advice below concerning interpretation of statistics.
“Lies, damned lies and statistics...”
A bad or dishonest statistician can "prove" just about anything. Let’s run through a few of basic rules concerning statistical surveys, using circumcision surveys as examples where appropriate. A failure to satisfy all the criteria mentioned below will not totally invalidate the results of any particular survey, but any deficiency should make readers wary of the results.
Reputable researchers involved with issues such as circumcision will, in the main, be working within an academic environment such as a University Medical School or Teaching Hospital. For many years all such establishments have had Research Ethics Committees that scrutinise research projects to ensure protection of the dignity, rights, safety and wellbeing of research participants. Some issues, particularly confidentiality, can crop up under more than one of these four headings. Approval of the study by such a committee should be stated in the report.
Random (or sufficiently pseudo-random) sampling
It is pointless carrying out a survey if the people responding to it do not constitute a reasonable cross-section of the population as a whole. Entire books have been devoted to this subject alone; a full description of sampling techniques is way beyond the scope of this article.
A particular issue to watch in the context of circumcision surveys is self-selection of respondents. If committed proponents and/or opponents submit a completed questionnaire but those who are ambivalent do not, a true picture of opinion will not be obtained. A questionnaire has to motivate those in the middle ground of opinion, otherwise only the extremes will show.
Separately, there can be a problem with activist groups (of either persuasion) mobilising their supporters with the deliberate intent of drowning out the contrary opinion. This latter point is a particular problem with online surveys, which are vulnerable in several respects to biased recruitment of voters.
The larger the sample, the less likely it is that distortions will result from inadequate sampling technique. Professional pollsters like to see a minimum of 1000 valid responses from respondents who are not self-selected. That’s not to say that smaller samples are of no use, merely that the probability of drawing a false conclusion rises as the sample size either shrinks or becomes less random.
Aside from sampling errors, there are other ways in which a poorly constructed survey can distort its own results. A classic example is the offering of multiple choice responses that omit some of the possibilities, forcing a reply that is not an expression of the respondent’s true opinion. Whenever the multiple choice format is used, a get-out option such as "No opinion", "Don’t know", "I don’t engage in that activity" or "Other, specify..." should normally be included.
Separately, a whole raft of bias issues can arise if respondents have access to the results before they themselves express their opinion. Again this is a major problem with many online surveys.
Another bias can arise through lack of confidentiality. "His and Hers" responses should be avoided when questionnaires are being designed. Example: A woman married to a man circumcised at birth may not want to admit to sexual experience with an uncircumcised male partner.
Security against multiple voting
Rather obviously, each participant should only be able to submit one response. In the absence of effective safeguards against multiple voting, a handful of activists can wreck a survey in no time at all.
Security against casual and ill-considered responses
Not everyone takes seriously the matter of submitting an accurate response to a research questionnaire. They may be in a silly mood, drunk, tired, bored, be distracted, have a headache, not understand the question or whatever. A well-designed survey will include less-than-obvious consistency checks intended to weed out misleading responses. It is perfectly legitimate for the researcher to discard responses that fail these consistency checks, but if this does happen the proportion thus discarded must be disclosed in the report. Too many discards indicate a poorly designed questionnaire.
Conclusions that sensibly interpret the results
A well-written survey report should state the results separately from the conclusions drawn from those results. If only the conclusions are stated, the report should be classified as an Abstract, not as a full report of the survey. An excellent example of possible misinterpretation of results is to be found on our web page about The People’s Republic of China
. There, we note findings that in the province of Chongqing 17.28% of boys aged 11 to 18 years are circumcised. Yet in the same geographical area none in the age range birth to 2 years are circumcised. This does not mean that the Chinese abandoned infant circumcision at some time in the past ten years! The correct interpetation, discernible from the detail of the research, is that circumcision is often recommended for boys deemed to be suffering from phimosis.
Beware also of errors involving correlation. Correlation does not
, of itself, prove a causal relationship. However, absence of correlation does disprove a causal relationship
Most professional publications, upon receiving a manuscript for publication, submit the copy to an expert in the same field for comment and opinion. It is not uncommon for research papers to be revised on the basis of this process of scrutiny, which is called "Peer Review". Be wary if peer review is absent. Also be wary if the reviewer is too closely associated with the researcher.
Declaration of competing interests or affiliations
Anything that might be perceived by others as a potential influence on the motivation for conducting the survey, on its results or on its conclusions should be declared. An example might be a survey about circumcision being sponsored by a clamp manufacturer. If there are no such declarations to be made, that too should be stated.
The following resources were used in the preparation of this web page:
To return to the previous page that you were viewing, please use the BACK button of your browser. Alternatively, choose another topic from the navigation panel below.
Copyright © 1992 - 2013, All Rights Reserved CIRCLIST.