STATISTICAL FALLACIES discussed in the book

"Almost all of us have legs more than the world average"

"There is a strong association between dying and being in bed" - so avoid bed to prolong life!!

"Statistics is like a bikini - what it reveals is interesting, what it conceals is vital"

"Head in an oven and feet in a freezer, and the person is comfortable, ON AVERAGE"

1. Problems with the sample

Biased sample (survivors, volunteers, clinic subjects, publication bias, inadequate specification of sampling method, abrupt series)

Inadequate sample (inadequate sample size, problems in calculation of size)

Incomparable groups (differential group composition, differential definitions, differential compliance, variable periods of exposure, improper denominator) 

Mixing of distinct groups (effect on regression, effect on shape of the distribution, lack of intra-group homogeneity)

2. Errors in presentation of findings

Misuse of perecentages and means (misuse of percentage, misuse of means, unnecessary decimals)

Problems in reporting (incomplete reporting, unspecified filtering, over-reporting, self-reporting vs. objective measurement, misuse of graphs)

3. Inadequate analysis

Ignoring reality (looking for linearity, overlooking assumptions, problems with area under the concentration curve, anamolous person-years)

General statistical problems (categorizing continuous variable, problems with assessment of Gaussianity, inaapropriate use of parametric tests, problems with nonparametric tests, CI vs. tests, inappropriate choice of effect measure, inappropriate analysis of serial data)

Mean or proportion (different conclusions)

Forgetting baseline values

Misuse of statistical packages (over-analysis, data dredging, quantitative nalysis of codes, soft data vs. hard data)

4. Misinterpretation

Misuse of P-values (magic threshold 0.05, one-tail or two-tail P-values, multiple comparisons, dramatic P-values, P-values for nonrandom sample, assessment of 'normal condition' invlving several parameters, inappropriate conclusion from large P-values)

Coorelation vs. cause-effect relationship (criteria for cause -effect, other considerations)

Sundry issues (diagnostic test is only an additional adjunct, medical significance vs. statistical significance, misinterpretation of standard error of p, univariate analysis but multivariate conclusions, limitation of relative reisk)

Make a free website with Yola