A New View of Statistics

You've got the idea!

1.

• a statistic NO
• a histogram YES
• a scatter plot YES
• a stem and leaf plot YES

2.

• for simpletons NO
• presented in stem and leaf plots NO
• things like correlations NO
• things like standard deviations YES

3.

• Use it do show spread. NO
• Use it for normally distributed data. NO NOT USUALLY
• Cross it against oncoming traffic. CAREFULLY!
• Indicate the middle of some data. YES

4.

• standard deviation YES
• root mean square errors YES
• percentile ranges YES
• polyunsaturated margarine SORT OF.

5.

• A NO
• B NO
• C YES
• D NO

6.

• a risk of matrimony NO
• an outcome statistic YES
• a relative of the odds ratio YES
• a relative frequency YES

7.

• p values NO
• standard errors of the mean DEFINITELY NOT!
• percentages of the mean NO
• standard deviations YES

8.

• describes loss of precision. NO
• describes factor analysis. YES
• is an example of ANOVA. NO
• is a weight-loss program. NO, BUT...

9.

• It impacts most on descriptive studies. NO
• It can be expressed as an ICC. YES
• It can be expressed as a CV. YES
• It is quantified by 2-way ANOVA. YES

10.

• It impacts most on descriptive studies. YES
• It is the correlation between true and observed values. YES
• A valid measure must be reliable. YES
• A reliable measure must be valid. NO

11.

• height = 175 ± 6 cm CORRECT
• VO2peak = 67 ± 5.1 ml/min/kg INCORRECT
• ICC = 0.87 CORRECT
• CV = 1.4% CORRECT

12.

• are a new form of sprint training. NOT A BAD IDEA...
• are calculated routinely by most stats packages. NO, CURSE THEM
• define the likely range of a population value. YES
• are inferior to p values for indicating magnitude of outcomes. OF COURSE NOT

13.

• The correlation is significant. NO
• The true value of the correlation is likely to be 0.45. NO
• More subjects should be tested. YES
• A type II error has occurred. NO (the population value might be insubstantial)

14.

• One-tailed tests are sometimes justified. NOT IN MY BOOK
• Test statistics should always be shown. NO POINT
• Chi-squared is a common test statistic. YES
• P = 0.06 means there is no effect. ABSOLUTELY WRONG

15.

• hardly ever NO
• about one time in 100 NO
• about one time in 20 NO (you'd get this if the population correlation was zero, not 0.70)
• almost always. YES (the only possible correct answer)

16.

• Differences between all groups are substantial. YES
• The data should be analyzed by repeated-measures ANOVA. NO
• Log transformation appears to be necessary before analysis. NO (rank transform, yes)
• Runners are lazier than cyclists. NO (running is harder than cycling)

17.

• is an example of an ordinal variable. YES
• has a behavior problem when it comes to residuals. YES
• should be analyzed by logistic regression. YES
• can be analyzed by ANOVA. NO

18.

• if the values are too big. NO
• if the residuals (error) get bigger for bigger values of the variable. YES
• if you don't get statistical significance. NO
• if non-parametric tests are inappropriate. NO

19.

• are parametric tests in disguise. YES
• involve rank transformation of the dependent variable. YES
• work for grossly non-normal data. YES
• should be attempted if parametric tests give p > 0.05. NO (but that's what some people do!)

20.

• unpaired t test. NO
• ANCOVA. YES
• ANOVA. NO
• MANOVA. NO

21.

• Use it to fit curves as well as straight lines. YES (polynomials)
• Use it to control for the effect of numeric variables. YES
• It gives misleading results for highly correlated independent variables. YES (neither appear to contribute in the presence of the other)
• Use it to fit multiple straight lines with several groups. NO

22.

• are used in descriptive studies. NO
• can be analyzed by modeling variances. YES
• are used when you have to repeat a failed test. NO
• are straightforward to analyze with stats programs. NO!

23.

• Initial randomization to the two groups was poor. NO. (How can you randomize sex!?)
• There is one between- and one within-subject factor. YES
• The time effect in the model is substantial. NO (time*sex is)
• The time effect in the model is significant. NO (need sample sizes to tell)

24.

• Sample size is proportional to (1 - r), where r=reliability correlation. YES
• Controlled studies need 4x as many subjects as uncontrolled studies. YES
• Get sample size "on the fly" by testing until you get an acceptable confidence interval. YES
• None of the above. NO

25.

• depends on the size of your research grant. NO/YES
• is inversely proportional to the square of the validities of your measures. YES
• is a function of the largest effect you want to detect. NO (the smallest...)
• depends on how many student researchers you have on the project. NO/YES

26.

• from now on you will show as few numbers as possible. YES
• statistical modeling is no substitute for knowing your data. YES
• it's important to play with stats programs. YES
• from now on you will test rather that estimate. NO!