prev

next

out of 54

View

214Download

0

Embed Size (px)

Stump the Statistician

Miscellaneous Design, SAS questions

Interactions and means comparisons

1.In a factorial or split plot analysis when there is a significant interaction between the two factors, how do I proceed to do treatment mean comparisons?

Interactions and means comparisons

1.In a factorial or split plot analysis when there is a significant interaction between the two factors, how do I proceed to do treatment mean comparisons?

Is this type of comparison meaningful?

Interactions and means comparisons

1.In a factorial or split plot analysis when there is a significant interaction between the two factors, how do I proceed to do treatment mean comparisons?

Is this type of comparison meaningful?

Nontrivial averaging over other factors.

Interactions and means comparisons

Is this type of comparison meaningful?

Nontrivial averaging over other factors.

Stratified main effects instead? (May require different error terms.)

Testing Assumptions

2.Which are the best tests for the assumptions of analysis of variance?

Testing Assumptions

2.Which are the best tests for the assumptions of analysis of variance?

Favorites, perhaps?

Testing Assumptions

2.Which are the best tests for the assumptions of analysis of variance?

Favorites, perhaps?

Normality: I like Wilk-Shapiro.

Testing Assumptions

2.Which are the best tests for the assumptions of analysis of variance?

Favorites, perhaps?

Normality: I like Wilk-Shapiro.

Good power against violations that give ANOVA fits.

Testing Assumptions

2.Which are the best tests for the assumptions of analysis of variance?

Favorites, perhaps?

Normality: I like Wilk-Shapiro.

Good power against violations that give ANOVA fits.

Not as nonspecific as Kolmogorov-Smirnov.

Testing Assumptions

2.Which are the best tests for the assumptions of analysis of variance?

Favorites, perhaps?

Normality: I like Wilk-Shapiro.

Good power against violations that give ANOVA fits.

Not as nonspecific as Kolmogorov-Smirnov.

Homoscedasticity: I like Levene.

Testing Assumptions

2.Which are the best tests for the assumptions of analysis of variance?

Favorites, perhaps?

Normality: I like Wilk-Shapiro.

Good power against violations that give ANOVA fits.

Not as nonspecific as Kolmogorov-Smirnov.

Homoscedasticity: I like Levene.

redo

of ANOVA on |ri

| (or ri2).

Testing Assumptions

2.Which are the best tests for the assumptions of analysis of variance?

Favorites, perhaps?

Normality: I like Wilk-Shapiro.

Good power against violations that give ANOVA fits.

Not as nonspecific as Kolmogorov-Smirnov.

Homoscedasticity: I like Levene.

redo

of ANOVA on |ri

| (or ri2).

If there are problems, points to a solution.

Outliers

3.Which is/are the best test(s) for identifying outliers?

Outliers

3.Which is/are the best test(s) for identifying outliers?

What is

truth?

Pontius Pilate

Outliers

3.Which is/are the best test(s) for identifying outliers?

What is

truth?

Pontius Pilate

Removing outliers changes your hypotheses.

Outliers

3.Which is/are the best test(s) for identifying outliers?

What is

truth?

Pontius Pilate

Removing outliers changes your hypotheses.

Correct errors first

Outliers

3.Which is/are the best test(s) for identifying outliers?

What is

truth?

Pontius Pilate

Removing outliers changes your hypotheses.

Correct errors first

Remove documentable outliers (no criterion).

Outliers

3.Which is/are the best test(s) for identifying outliers?

What is

truth?

Pontius Pilate

Removing outliers changes your hypotheses.

Correct errors first

Remove documentable outliers (no criterion).

Removal based on test results as a final resort.

Transformations

4.How can I tell if a particular transformation is statistically appropriate?

Transformations

4.How can I tell if a particular transformation is statistically appropriate?

Biological considerations.

Transformations

4.How can I tell if a particular transformation is statistically appropriate?

Biological considerations.

Hypotheses change again.

Transformations

4.How can I tell if a particular transformation is statistically appropriate?

Biological considerations.

Hypotheses change again.

Interpretation of outliers.

Transformations

4.How can I tell if a particular transformation is statistically appropriate?

Biological considerations.

Hypotheses change again.

Interpretation of outliers.

Address normality first, constant variance 2nd.

Transformations

4.How can I tell if a particular transformation is statistically appropriate?

Biological considerations.

Hypotheses change again.

Interpretation of outliers.

Address normality first, constant variance 2nd.

Dont believe what you learned in school.

E.G., re: arcsine square root transformation.

Sample size calculation

5.If I have a general idea of what the coefficient of variation of a variable is likely to be prior to the start of a field trial, how do I determine how many replicate blocks I should test to have a reasonable shot at getting significance?

Sample size calculation

5.If I have a general idea of what the coefficient of variation of a variable is likely to be prior to the start of a field trial, how do I determine how many replicate blocks I should test to have a reasonable shot at getting significance?

This assumes proportional effects / log transformation.

Sample size calculation

5.If I have a general idea of what the coefficient of variation of a variable is likely to be prior to the start of a field trial, how do I determine how many replicate blocks I should test to have a reasonable shot at getting significance?

This assumes proportional effects / log transformation.

Tools include Russ Lenths

applets.

Polynomial curve fitting

6.On a quantitative variable (fertilizer rate, PGR rate, etc.) I get significant linear and quadratic effects; what are the rules for deciding which response form is the 'real' one?

Polynomial curve fitting

6.On a quantitative variable (fertilizer rate, PGR rate, etc.) I get significant linear and quadratic effects; what are the rules for deciding which response form is the 'real' one?

Why polynomial in the first place? (Taylor)

Polynomial curve fitting

6.On a quantitative variable (fertilizer rate, PGR rate, etc.) I get significant linear and quadratic effects; what are the rules for deciding which response form is the 'real' one?

Why polynomial in the first place? (Taylor)

Does the biology help you answer this?

Polynomial curve fitting

Why polynomial in the first place? (Taylor)

Does the biology help you answer this?

Transform if necessary.

Polynomial curve fitting

Why polynomial in the first place? (Taylor)

Does the biology help you answer this?

Transform if necessary.

Nonlinear tools (e.g., GAM)

Changing your ANOVA model

7.On a multiple factor test I run the anova with the block and all interaction terms, all

of which turn out to be non-significant. The main effects are not quite significant at p

Changing your ANOVA model

7. Is changing original model kosher?

Probably not.

Changing your ANOVA model

7. Is changing original model kosher?

Probably not.

Is .05 sacred? Is .049 all that different from .051?

Changing your ANOVA model

7. Is changing original model kosher?

Probably not.

Is .05 sacred? Is .049 all that different from .051?

Why did you block in first place?

Changing your ANOVA model

7. Is changing original model kosher?

Probably not.

Is .05 sacred? Is .049 all that different from .051?

Why did you block in first place?

Opportunity to correct stupid mistakes?

Post hoc comparisons

8.Is it appropriate to run contrasts on groups of treatments if the overall anova

was not

significant? How about LSD analysis?

Post hoc comparisons

8.Is it appropriate to run contrasts on groups of treatments if the overall anova

was not

significant? How about LSD analysis?

LSD versus PLSD.

Post hoc comparisons

8.Is it appropriate to run contrasts on groups of treatments if the overall anova

was not

significant? How about LSD analysis?

LSD ve