A PRIORI OR PLANNED CONTRASTS MULTIPLE COMPARISON TESTS
![A PRIORI OR PLANNED CONTRASTS MULTIPLE COMPARISON TESTS A PRIORI OR PLANNED CONTRASTS MULTIPLE COMPARISON TESTS](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-1.jpg)
![ANOVA �ANOVA is used to compare means. �However, if a difference is detected, and ANOVA �ANOVA is used to compare means. �However, if a difference is detected, and](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-2.jpg)
![PLANNED or A PRIORI CONTRASTS �A comparison between means identified as being of utmost PLANNED or A PRIORI CONTRASTS �A comparison between means identified as being of utmost](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-3.jpg)
![UNPLANNED or A POSTERIORI CONTRASTS �A form of “data dredging” or “data snooping”, where UNPLANNED or A POSTERIORI CONTRASTS �A form of “data dredging” or “data snooping”, where](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-4.jpg)
![PLANNED ORTHOGONAL AND NON-ORTHOGONAL CONTRASTS �Planned comparisons may be orthogonal or non-orthogonal. �Orthogonal: mutually PLANNED ORTHOGONAL AND NON-ORTHOGONAL CONTRASTS �Planned comparisons may be orthogonal or non-orthogonal. �Orthogonal: mutually](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-5.jpg)
![ORTHOGONAL CONTRASTS �Limited number of contrasts can be made, simultaneously. �Any set of contrasts ORTHOGONAL CONTRASTS �Limited number of contrasts can be made, simultaneously. �Any set of contrasts](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-6.jpg)
![ORTHOGONAL CONTRASTS �For Example: k=4 means. Therefore, you can make 3 (i. e. : ORTHOGONAL CONTRASTS �For Example: k=4 means. Therefore, you can make 3 (i. e. :](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-7.jpg)
![HOW DO YOU KNOW IF A SET OF CONTRASTS IS ORTHOGONAL? �∑cijci’j=0 where the HOW DO YOU KNOW IF A SET OF CONTRASTS IS ORTHOGONAL? �∑cijci’j=0 where the](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-8.jpg)
![HOW DO YOU KNOW IF A SET OF CONTRASTS IS ORTHOGONAL? �For Example: Set HOW DO YOU KNOW IF A SET OF CONTRASTS IS ORTHOGONAL? �For Example: Set](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-9.jpg)
![ORTHOGONAL CONTRASTS �There always k-1 non-redundant questions that can be answered. �An experimenter may ORTHOGONAL CONTRASTS �There always k-1 non-redundant questions that can be answered. �An experimenter may](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-10.jpg)
![PLANNED COMPARISONS USING A t STATISTIC �A planned comparison addresses the null hypothesis that PLANNED COMPARISONS USING A t STATISTIC �A planned comparison addresses the null hypothesis that](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-11.jpg)
![PLANNED COMPARISONS USING A t STATISTIC �Very similar to a two sample t-test, but PLANNED COMPARISONS USING A t STATISTIC �Very similar to a two sample t-test, but](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-12.jpg)
![PLANNED COMPARISONS USING A t STATISTIC �Evaluate just like any other t-test. �Look up PLANNED COMPARISONS USING A t STATISTIC �Evaluate just like any other t-test. �Look up](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-13.jpg)
![PLANNED COMPARISONS USING A t STATISTIC: NOTE �All of the t statistic calculations for PLANNED COMPARISONS USING A t STATISTIC: NOTE �All of the t statistic calculations for](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-14.jpg)
![PLANNED COMPARISONS USING AN F STATISTIC �You can also use an F statistic for PLANNED COMPARISONS USING AN F STATISTIC �You can also use an F statistic for](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-15.jpg)
![CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �A confidence interval is a way of expressing the CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �A confidence interval is a way of expressing the](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-16.jpg)
![CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �The values at the extremes are the 95% confidence CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �The values at the extremes are the 95% confidence](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-17.jpg)
![ADVANTAGES OF USING CONFIDENCE INTERVALS �When the data are presented this way, it is ADVANTAGES OF USING CONFIDENCE INTERVALS �When the data are presented this way, it is](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-18.jpg)
![CHOOSING A METHOD �Orthogonal tests can be done in either way. �Both methods make CHOOSING A METHOD �Orthogonal tests can be done in either way. �Both methods make](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-19.jpg)
![ASSUMPTIONS �Assumptions: The populations are approximately normally distributed. Their variances are homogenous. �The t ASSUMPTIONS �Assumptions: The populations are approximately normally distributed. Their variances are homogenous. �The t](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-20.jpg)
![HOW TO DEAL WITH VIOLATIONS OF ASSUMPTIONS �When population variances are unequal, you can HOW TO DEAL WITH VIOLATIONS OF ASSUMPTIONS �When population variances are unequal, you can](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-21.jpg)
![TYPE I ERRORS AND ORTHOGONAL CONTRASTS �For C independent contrasts at some level of TYPE I ERRORS AND ORTHOGONAL CONTRASTS �For C independent contrasts at some level of](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-22.jpg)
![A PRIORI NON-ORTHOGONAL CONTRASTS �Contrasts of interest that ARE NOT independent. �In order to A PRIORI NON-ORTHOGONAL CONTRASTS �Contrasts of interest that ARE NOT independent. �In order to](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-23.jpg)
![WHEN DO YOU DO THESE? �When contrasts are planned in advance. �They are relatively WHEN DO YOU DO THESE? �When contrasts are planned in advance. �They are relatively](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-24.jpg)
![DUNN’S MULTIPLE COMPARISON PROCEDURE �A. K. A. : Bonferoni t procedure. �Involves splitting up DUNN’S MULTIPLE COMPARISON PROCEDURE �A. K. A. : Bonferoni t procedure. �Involves splitting up](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-25.jpg)
![DUNN’S MULTIPLE COMPARISON PROCEDURE �If the consequences of making a Type 1 error are DUNN’S MULTIPLE COMPARISON PROCEDURE �If the consequences of making a Type 1 error are](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-26.jpg)
![DUNN’S MULTIPLE COMPARISON PROCEDURE �This procedure also involves the calculation of a t statistic DUNN’S MULTIPLE COMPARISON PROCEDURE �This procedure also involves the calculation of a t statistic](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-27.jpg)
![DUNN’S MULTIPLE COMPARISON PROCEDURE �However, you use a different table in order to look DUNN’S MULTIPLE COMPARISON PROCEDURE �However, you use a different table in order to look](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-28.jpg)
![DUNN’S MULTIPLE COMPARISON PROCEDURE: ONE-TAILED TESTS �The table also only shows the critical values DUNN’S MULTIPLE COMPARISON PROCEDURE: ONE-TAILED TESTS �The table also only shows the critical values](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-29.jpg)
![DUNN’S MULTIPLE COMPARISON PROCEDURE �Instead of calculating t. D for all contrasts of interest, DUNN’S MULTIPLE COMPARISON PROCEDURE �Instead of calculating t. D for all contrasts of interest,](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-30.jpg)
![DUNN’S MULTIPLE COMPARISON PROCEDURE For Example: You have 5 means (Y 1 through Y DUNN’S MULTIPLE COMPARISON PROCEDURE For Example: You have 5 means (Y 1 through Y](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-31.jpg)
![DUNN-SIDAK PROCEDURE �A modification of the Dunn procedure. t statistic (t. DS) and critical DUNN-SIDAK PROCEDURE �A modification of the Dunn procedure. t statistic (t. DS) and critical](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-32.jpg)
![DUNN-SIDAK PROCEDURE �However, it is not easy to allocate the total value of α DUNN-SIDAK PROCEDURE �However, it is not easy to allocate the total value of α](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-33.jpg)
![DUNNETT’S TEST �For contrasts involving a control mean. �Also uses a t statistic (t. DUNNETT’S TEST �For contrasts involving a control mean. �Also uses a t statistic (t.](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-34.jpg)
![CHOOSING A PROCEDURE : A PRIORI NON-ORTHOGONAL TESTS �Often, the use of more than CHOOSING A PROCEDURE : A PRIORI NON-ORTHOGONAL TESTS �Often, the use of more than](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-35.jpg)
![A PRIORI ORTHOGONAL and NONORTHOGONAL CONTRASTS �The advantage of being able to make all A PRIORI ORTHOGONAL and NONORTHOGONAL CONTRASTS �The advantage of being able to make all](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-36.jpg)
![A PRIORI and A POSTERIORI NONORTHOGONAL CONTRASTS �When you have a large number of A PRIORI and A POSTERIORI NONORTHOGONAL CONTRASTS �When you have a large number of](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-37.jpg)
![A POSTERIORI CONTRASTS �There are many kinds, all offering different degrees of protection from A POSTERIORI CONTRASTS �There are many kinds, all offering different degrees of protection from](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-38.jpg)
![A POSTERIORI CONTRASTS �Most are good for doing all possible pair- wise comparisons between A POSTERIORI CONTRASTS �Most are good for doing all possible pair- wise comparisons between](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-39.jpg)
![CHOOSING AN APPROPRIATE TEST PROCEDURE �Trade-off between power and the probability of making Type CHOOSING AN APPROPRIATE TEST PROCEDURE �Trade-off between power and the probability of making Type](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-40.jpg)
![DOING PLANNED CONTRASTS USING SAS �You have a data set with one dependent variable DOING PLANNED CONTRASTS USING SAS �You have a data set with one dependent variable](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-41.jpg)
![ARE THEY ORTHOGONAL? SET 1 cijci’j ∑cijci’j COEFFICIENTS CONTRASTS c 1 c 2 c ARE THEY ORTHOGONAL? SET 1 cijci’j ∑cijci’j COEFFICIENTS CONTRASTS c 1 c 2 c](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-42.jpg)
![DOING PLANNED ORTHOGONAL CONTRASTS USING SAS � SAS INPUT: dataset; Give your dataset an DOING PLANNED ORTHOGONAL CONTRASTS USING SAS � SAS INPUT: dataset; Give your dataset an](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-43.jpg)
![DOING PLANNED ORTHOGONAL CONTRASTS USING SAS ANOVA DOING PLANNED ORTHOGONAL CONTRASTS USING SAS ANOVA](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-44.jpg)
![RECOMMENDED READING �Kirk RE. 1982. Experimental design: procedures for the behavioural sciences. Second ed. RECOMMENDED READING �Kirk RE. 1982. Experimental design: procedures for the behavioural sciences. Second ed.](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-45.jpg)
![A PRIORI OR PLANNED CONTRASTS THE END A PRIORI OR PLANNED CONTRASTS THE END](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-46.jpg)
![EXTRA SLIDES EXTRA SLIDES](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-47.jpg)
![PLANNED COMPARISONS USING A t STATISTIC �t = ∑cj. Yj/ √MSerror∑cj/nj Where c is PLANNED COMPARISONS USING A t STATISTIC �t = ∑cj. Yj/ √MSerror∑cj/nj Where c is](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-48.jpg)
![PLANNED COMPARISONS USING A t STATISTIC �For Example, you want to compare 2 means: PLANNED COMPARISONS USING A t STATISTIC �For Example, you want to compare 2 means:](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-49.jpg)
![CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �(Y 1 -Y 2) – tα/2, v(SE) ≤ ψ CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �(Y 1 -Y 2) – tα/2, v(SE) ≤ ψ](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-50.jpg)
![TYPE I ERRORS AND ORTHOGONAL CONTRASTS �As the number of independent tests increases, so TYPE I ERRORS AND ORTHOGONAL CONTRASTS �As the number of independent tests increases, so](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-51.jpg)
- Slides: 51
![A PRIORI OR PLANNED CONTRASTS MULTIPLE COMPARISON TESTS A PRIORI OR PLANNED CONTRASTS MULTIPLE COMPARISON TESTS](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-1.jpg)
A PRIORI OR PLANNED CONTRASTS MULTIPLE COMPARISON TESTS
![ANOVA ANOVA is used to compare means However if a difference is detected and ANOVA �ANOVA is used to compare means. �However, if a difference is detected, and](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-2.jpg)
ANOVA �ANOVA is used to compare means. �However, if a difference is detected, and more than two means are being compared, ANOVA cannot tell you where the difference lies. �In order to figure out which means differ, you can do a series of tests: Planned or unplanned comparisons of means.
![PLANNED or A PRIORI CONTRASTS A comparison between means identified as being of utmost PLANNED or A PRIORI CONTRASTS �A comparison between means identified as being of utmost](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-3.jpg)
PLANNED or A PRIORI CONTRASTS �A comparison between means identified as being of utmost interest during the design of a study, prior to data collection. �You can only do one or a very small number of planned comparisons, otherwise you risk inflating the Type 1 error rate. �You do not need to perform an ANOVA first.
![UNPLANNED or A POSTERIORI CONTRASTS A form of data dredging or data snooping where UNPLANNED or A POSTERIORI CONTRASTS �A form of “data dredging” or “data snooping”, where](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-4.jpg)
UNPLANNED or A POSTERIORI CONTRASTS �A form of “data dredging” or “data snooping”, where you may perform comparisons between all potential pairs of means in order to figure out where the difference(s) lie. �No prior justification for comparisons. �Increased risk of committing a Type 1 error. �The probability of making at least one type 1 error is not greater than α= 0. 05.
![PLANNED ORTHOGONAL AND NONORTHOGONAL CONTRASTS Planned comparisons may be orthogonal or nonorthogonal Orthogonal mutually PLANNED ORTHOGONAL AND NON-ORTHOGONAL CONTRASTS �Planned comparisons may be orthogonal or non-orthogonal. �Orthogonal: mutually](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-5.jpg)
PLANNED ORTHOGONAL AND NON-ORTHOGONAL CONTRASTS �Planned comparisons may be orthogonal or non-orthogonal. �Orthogonal: mutually non-redundant and uncorrelated contrasts (i. e. : independent). �Non-Orthogonal: Not independent. �For example: 4 means: Y 1 , Y 2, Y 3, and Y 4 Orthogonal: Y 1 - Y 2 and Y 3 - Y 4 Non-Orthogonal: Y 1 -Y 2 and Y 2 -Y 3
![ORTHOGONAL CONTRASTS Limited number of contrasts can be made simultaneously Any set of contrasts ORTHOGONAL CONTRASTS �Limited number of contrasts can be made, simultaneously. �Any set of contrasts](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-6.jpg)
ORTHOGONAL CONTRASTS �Limited number of contrasts can be made, simultaneously. �Any set of contrasts may have k-1 number of contrasts.
![ORTHOGONAL CONTRASTS For Example k4 means Therefore you can make 3 i e ORTHOGONAL CONTRASTS �For Example: k=4 means. Therefore, you can make 3 (i. e. :](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-7.jpg)
ORTHOGONAL CONTRASTS �For Example: k=4 means. Therefore, you can make 3 (i. e. : 4 -1) orthogonal contrasts at once.
![HOW DO YOU KNOW IF A SET OF CONTRASTS IS ORTHOGONAL cijcij0 where the HOW DO YOU KNOW IF A SET OF CONTRASTS IS ORTHOGONAL? �∑cijci’j=0 where the](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-8.jpg)
HOW DO YOU KNOW IF A SET OF CONTRASTS IS ORTHOGONAL? �∑cijci’j=0 where the c’s are the particular coefficients associated with each of the means and the i indicates the particular comparison to which you are referring. �Multiply all of the coefficients for each particular mean together across all comparisons. �Then add them up! �If that sum is equal to zero, then the comparisons that you have in your set may be considered orthogonal.
![HOW DO YOU KNOW IF A SET OF CONTRASTS IS ORTHOGONAL For Example Set HOW DO YOU KNOW IF A SET OF CONTRASTS IS ORTHOGONAL? �For Example: Set](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-9.jpg)
HOW DO YOU KNOW IF A SET OF CONTRASTS IS ORTHOGONAL? �For Example: Set c 1 1 ½ ½ 1 cijci’j ∑cijci’j Coefficients c 2 c 3 -1 1 ½ -½ -½ -½ Contrasts c 4 -1 -½ ½ Y 1 -Y 2 Y 3 -Y 4 (Y 1+Y 2)/2 –(Y 3+Y 4)/2 0 (c 1)Y 1 + (c 2)Y 2 = Y 1 – Y 2 Therefore c 1 = 1 and c 2 = -1 because (1)Y 1 + (-1)Y 2 = Y 1 -Y 2 (After Kirk 1982)
![ORTHOGONAL CONTRASTS There always k1 nonredundant questions that can be answered An experimenter may ORTHOGONAL CONTRASTS �There always k-1 non-redundant questions that can be answered. �An experimenter may](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-10.jpg)
ORTHOGONAL CONTRASTS �There always k-1 non-redundant questions that can be answered. �An experimenter may not be interested in asking all of said questions, however.
![PLANNED COMPARISONS USING A t STATISTIC A planned comparison addresses the null hypothesis that PLANNED COMPARISONS USING A t STATISTIC �A planned comparison addresses the null hypothesis that](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-11.jpg)
PLANNED COMPARISONS USING A t STATISTIC �A planned comparison addresses the null hypothesis that all of your comparisons between means will be equal to zero. Ho=Y 1 -Y 2=0 Ho= Y 3 -Y 4=0 Ho= (Y 1+Y 2)/2 –(Y 3+Y 4)/2 �These types of hypotheses can be tested using a t statistic.
![PLANNED COMPARISONS USING A t STATISTIC Very similar to a two sample ttest but PLANNED COMPARISONS USING A t STATISTIC �Very similar to a two sample t-test, but](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-12.jpg)
PLANNED COMPARISONS USING A t STATISTIC �Very similar to a two sample t-test, but the standard error is calculated differently. �Specifically, planned comparisons use the pooled sample variance (MSerror )based on all k groups (and the corresponding error degrees of freedom) rather than that based only on the two groups being compared. �This step increases precision and power.
![PLANNED COMPARISONS USING A t STATISTIC Evaluate just like any other ttest Look up PLANNED COMPARISONS USING A t STATISTIC �Evaluate just like any other t-test. �Look up](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-13.jpg)
PLANNED COMPARISONS USING A t STATISTIC �Evaluate just like any other t-test. �Look up the critical value for t in the same table. �If the absolute value of your calculated t statistic exceeds the critical value, the null hypothesis is rejected.
![PLANNED COMPARISONS USING A t STATISTIC NOTE All of the t statistic calculations for PLANNED COMPARISONS USING A t STATISTIC: NOTE �All of the t statistic calculations for](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-14.jpg)
PLANNED COMPARISONS USING A t STATISTIC: NOTE �All of the t statistic calculations for all of the comparisons in a particular set will use the same MSerror. �Thus, the tests themselves are not statistically independent, even though the comparisons that you are making are. �However, it has been shown that, if you have a sufficiently large number of degrees of freedom (40+), this shouldn’t matter. (Norton and Bulgren, as cited by Kirk, 1982)
![PLANNED COMPARISONS USING AN F STATISTIC You can also use an F statistic for PLANNED COMPARISONS USING AN F STATISTIC �You can also use an F statistic for](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-15.jpg)
PLANNED COMPARISONS USING AN F STATISTIC �You can also use an F statistic for these tests, because t 2 = F. �Different books prefer different methods. �The book I liked most used the t statistic, so that’s what I’m going to use throughout. �SAS uses F, however.
![CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS A confidence interval is a way of expressing the CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �A confidence interval is a way of expressing the](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-16.jpg)
CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �A confidence interval is a way of expressing the precision of an estimate of a parameter. �Here, the parameter that we are estimating is the value of the particular contrast that we are making. �So, the actual value of the comparison (ψ) should be somewhere between the two extremes of the confidence interval.
![CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS The values at the extremes are the 95 confidence CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �The values at the extremes are the 95% confidence](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-17.jpg)
CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �The values at the extremes are the 95% confidence limits. �With them, you can say that you are 95% confident that the true value of the comparison lies between those two values. �If the confidence interval does not include zero, then you can conclude that the null hypothesis can be rejected.
![ADVANTAGES OF USING CONFIDENCE INTERVALS When the data are presented this way it is ADVANTAGES OF USING CONFIDENCE INTERVALS �When the data are presented this way, it is](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-18.jpg)
ADVANTAGES OF USING CONFIDENCE INTERVALS �When the data are presented this way, it is possible for the experimenter to consider all possible null hypotheses – not just the one that states that the comparison in question will equal 0. �If any hypothesized value lies outside of the 95% confidence interval, it can be rejected.
![CHOOSING A METHOD Orthogonal tests can be done in either way Both methods make CHOOSING A METHOD �Orthogonal tests can be done in either way. �Both methods make](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-19.jpg)
CHOOSING A METHOD �Orthogonal tests can be done in either way. �Both methods make the same assumptions and are equally significant.
![ASSUMPTIONS Assumptions The populations are approximately normally distributed Their variances are homogenous The t ASSUMPTIONS �Assumptions: The populations are approximately normally distributed. Their variances are homogenous. �The t](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-20.jpg)
ASSUMPTIONS �Assumptions: The populations are approximately normally distributed. Their variances are homogenous. �The t statistic is relatively robust to violations of these assumptions when the number of observations for each sample are equal. �However, when the sample sizes are not equal, the t statistic is not robust to the heterogeneity of variances.
![HOW TO DEAL WITH VIOLATIONS OF ASSUMPTIONS When population variances are unequal you can HOW TO DEAL WITH VIOLATIONS OF ASSUMPTIONS �When population variances are unequal, you can](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-21.jpg)
HOW TO DEAL WITH VIOLATIONS OF ASSUMPTIONS �When population variances are unequal, you can replace the pooled estimator of variance, MSerror, with individual variance estimators for the means that you are comparing. �There a number of possible procedures that can be used when the variance between populations is heterogeneous: Cochran and Cox Welch Dixon, Massey, Satterthwaite and Smith
![TYPE I ERRORS AND ORTHOGONAL CONTRASTS For C independent contrasts at some level of TYPE I ERRORS AND ORTHOGONAL CONTRASTS �For C independent contrasts at some level of](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-22.jpg)
TYPE I ERRORS AND ORTHOGONAL CONTRASTS �For C independent contrasts at some level of significance (α), the probability of making one or more Type 1 errors is equal to: 1 -(1 -α)C �As the number of independent tests increases, so does the probability of committing a Type 1 error. �This problem can be reduced (but not eliminated) by restricting the use of multiple t-tests to a priori orthogonal contrasts.
![A PRIORI NONORTHOGONAL CONTRASTS Contrasts of interest that ARE NOT independent In order to A PRIORI NON-ORTHOGONAL CONTRASTS �Contrasts of interest that ARE NOT independent. �In order to](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-23.jpg)
A PRIORI NON-ORTHOGONAL CONTRASTS �Contrasts of interest that ARE NOT independent. �In order to reduce the probability of making a Type 1 error, the significance level (α) is set for the whole family of comparisons that is being made, as opposed to for each individual comparison. �For Example: Entire value of α for all comparisons combined is 0. 05. The value for each individual comparison would thus be less than that.
![WHEN DO YOU DO THESE When contrasts are planned in advance They are relatively WHEN DO YOU DO THESE? �When contrasts are planned in advance. �They are relatively](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-24.jpg)
WHEN DO YOU DO THESE? �When contrasts are planned in advance. �They are relatively few in number. � BUT the comparisons are non-orthogonal (they are not independent). �i. e. : When one wants to contrast a control group mean with experimental group means.
![DUNNS MULTIPLE COMPARISON PROCEDURE A K A Bonferoni t procedure Involves splitting up DUNN’S MULTIPLE COMPARISON PROCEDURE �A. K. A. : Bonferoni t procedure. �Involves splitting up](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-25.jpg)
DUNN’S MULTIPLE COMPARISON PROCEDURE �A. K. A. : Bonferoni t procedure. �Involves splitting up the value of α among a set of planned contrasts in an additive way. �For example: Total α = 0. 05, for all contrasts. One is doing 2 contrasts. α for each contrast could be 0. 025, if we wanted to divide up the α equally.
![DUNNS MULTIPLE COMPARISON PROCEDURE If the consequences of making a Type 1 error are DUNN’S MULTIPLE COMPARISON PROCEDURE �If the consequences of making a Type 1 error are](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-26.jpg)
DUNN’S MULTIPLE COMPARISON PROCEDURE �If the consequences of making a Type 1 error are not equally serious for all contrasts, then you may choose to divide α unequally across all of the possible comparisons in order to reflect that concern.
![DUNNS MULTIPLE COMPARISON PROCEDURE This procedure also involves the calculation of a t statistic DUNN’S MULTIPLE COMPARISON PROCEDURE �This procedure also involves the calculation of a t statistic](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-27.jpg)
DUNN’S MULTIPLE COMPARISON PROCEDURE �This procedure also involves the calculation of a t statistic (t. D). �The calculation involved in finding t. D is identical to that for determining t for orthogonal tests:
![DUNNS MULTIPLE COMPARISON PROCEDURE However you use a different table in order to look DUNN’S MULTIPLE COMPARISON PROCEDURE �However, you use a different table in order to look](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-28.jpg)
DUNN’S MULTIPLE COMPARISON PROCEDURE �However, you use a different table in order to look up the critical value (t. Dα; C, v ). Your total α value (not the value per comparison). Number of comparisons (C). And v, the number of degrees of freedom.
![DUNNS MULTIPLE COMPARISON PROCEDURE ONETAILED TESTS The table also only shows the critical values DUNN’S MULTIPLE COMPARISON PROCEDURE: ONE-TAILED TESTS �The table also only shows the critical values](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-29.jpg)
DUNN’S MULTIPLE COMPARISON PROCEDURE: ONE-TAILED TESTS �The table also only shows the critical values for two-tailed tests. �However, you can determine the approximate value of t. Dα; C, v for a onetailed test by using the following equation: t. Dα; C, v ≈ zα/C + (z 3α/C + zα/C)/4(v-2) ▪ Where the value of zα/C can be looked up in yet another table (“Areas under the Standard Normal Distribution”).
![DUNNS MULTIPLE COMPARISON PROCEDURE Instead of calculating t D for all contrasts of interest DUNN’S MULTIPLE COMPARISON PROCEDURE �Instead of calculating t. D for all contrasts of interest,](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-30.jpg)
DUNN’S MULTIPLE COMPARISON PROCEDURE �Instead of calculating t. D for all contrasts of interest, you can simply calculate the critical difference (ψD) that a particular comparison must exceed in order to be significant: ψD = t. Dα/2; C, v √(2 MSerror/n). �Then compare this critical difference value to the absolute values of the differences between the means that you compared. �If they exceed ψD, they are significant.
![DUNNS MULTIPLE COMPARISON PROCEDURE For Example You have 5 means Y 1 through Y DUNN’S MULTIPLE COMPARISON PROCEDURE For Example: You have 5 means (Y 1 through Y](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-31.jpg)
DUNN’S MULTIPLE COMPARISON PROCEDURE For Example: You have 5 means (Y 1 through Y 5). Those differences that exceed the calculated ΨD = 8. 45 value of ΨD (8. 45, in this case) are significant. Differences between means are: MEANS Y 1 Y 2 Y 3 Y 4 Y 5 - 12. 0 6. 7 10. 5 3. 6 - 5. 3 1. 5 8. 4 - 3. 8 3. 1 - 6. 9 (After Kirk 1982)
![DUNNSIDAK PROCEDURE A modification of the Dunn procedure t statistic t DS and critical DUNN-SIDAK PROCEDURE �A modification of the Dunn procedure. t statistic (t. DS) and critical](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-32.jpg)
DUNN-SIDAK PROCEDURE �A modification of the Dunn procedure. t statistic (t. DS) and critical difference (ψDS). �There isn’t much difference between the two procedures at α < 0. 01. �However, at increased values of α, this procedure is considered to be more powerful and more precise. �Calculations are the same for t and ψD. �Table is different.
![DUNNSIDAK PROCEDURE However it is not easy to allocate the total value of α DUNN-SIDAK PROCEDURE �However, it is not easy to allocate the total value of α](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-33.jpg)
DUNN-SIDAK PROCEDURE �However, it is not easy to allocate the total value of α unevenly across a particular set of comparisons. �This is because the values of α for each individual comparison are related multiplicatively, as opposed to additively. �Thus, you can’t simply add the α’s for each comparison together to get the total value of α for all contrasts combined.
![DUNNETTS TEST For contrasts involving a control mean Also uses a t statistic t DUNNETT’S TEST �For contrasts involving a control mean. �Also uses a t statistic (t.](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-34.jpg)
DUNNETT’S TEST �For contrasts involving a control mean. �Also uses a t statistic (t. D’) and critical difference (ψD’). �Calculations are the same for t and ψ. �Different table. �Instead of C, you use k, the number of means (including the control mean). �Note: unlike Dunn’s and Dunn-Sidak’s, Dunnet’s procedure is limited to k-1 non-orthogonal comparisons.
![CHOOSING A PROCEDURE A PRIORI NONORTHOGONAL TESTS Often the use of more than CHOOSING A PROCEDURE : A PRIORI NON-ORTHOGONAL TESTS �Often, the use of more than](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-35.jpg)
CHOOSING A PROCEDURE : A PRIORI NON-ORTHOGONAL TESTS �Often, the use of more than one procedure will appear to be appropriate. �In such cases, compute the critical difference (ψ) necessary to reject the null hypothesis for all of the possible procedures. �Use the one that gives the smallest critical difference (ψ) value.
![A PRIORI ORTHOGONAL and NONORTHOGONAL CONTRASTS The advantage of being able to make all A PRIORI ORTHOGONAL and NONORTHOGONAL CONTRASTS �The advantage of being able to make all](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-36.jpg)
A PRIORI ORTHOGONAL and NONORTHOGONAL CONTRASTS �The advantage of being able to make all planned contrasts, not just those that are orthogonal, is gained at the expense of an increase in the probability of making Type 2 errors.
![A PRIORI and A POSTERIORI NONORTHOGONAL CONTRASTS When you have a large number of A PRIORI and A POSTERIORI NONORTHOGONAL CONTRASTS �When you have a large number of](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-37.jpg)
A PRIORI and A POSTERIORI NONORTHOGONAL CONTRASTS �When you have a large number of means, but only comparatively very few contrasts, a priori non-orthogonal contrasts are better suited. �However, if you have relatively few means and a larger number of contrasts, you may want to consider doing an a posteriori test instead.
![A POSTERIORI CONTRASTS There are many kinds all offering different degrees of protection from A POSTERIORI CONTRASTS �There are many kinds, all offering different degrees of protection from](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-38.jpg)
A POSTERIORI CONTRASTS �There are many kinds, all offering different degrees of protection from Type 1 and Type 2 errors: Least Significant Difference (LSD) Test Tukey’s Honestly Significant Difference (HSD) Test Spjtotvoll and Stoline HSD Test Tukey-Kramer HSD Test Scheffé’s S Test Brown-Forsythe BF Procedure Newman-Keuls Test Duncan’s New Multiple Range Test
![A POSTERIORI CONTRASTS Most are good for doing all possible pair wise comparisons between A POSTERIORI CONTRASTS �Most are good for doing all possible pair- wise comparisons between](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-39.jpg)
A POSTERIORI CONTRASTS �Most are good for doing all possible pair- wise comparisons between means. �One (Scheffé’s method) allows you to evaluate all possible contrasts between means, whether they are pair-wise or not.
![CHOOSING AN APPROPRIATE TEST PROCEDURE Tradeoff between power and the probability of making Type CHOOSING AN APPROPRIATE TEST PROCEDURE �Trade-off between power and the probability of making Type](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-40.jpg)
CHOOSING AN APPROPRIATE TEST PROCEDURE �Trade-off between power and the probability of making Type 1 errors. �When a test is conservative, it is less likely that you will make a Type 1 error. �But it also would lack power, inflating the Type 2 error rate. �You will want to control the Type 1 error rate without loosing too much power. �Otherwise, you might reject differences between means that are actually significant.
![DOING PLANNED CONTRASTS USING SAS You have a data set with one dependent variable DOING PLANNED CONTRASTS USING SAS �You have a data set with one dependent variable](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-41.jpg)
DOING PLANNED CONTRASTS USING SAS �You have a data set with one dependent variable (y) and one independent variable (a). �(a) has 4 different treatment levels (1, 2, 3, and 4). �You want to do the following comparisons between treatment levels: 1)Y 1 -Y 2 2) Y 3 -Y 4 3) (Y 1 -Y 2)/2 - (Y 3 -Y 4)/2
![ARE THEY ORTHOGONAL SET 1 cijcij cijcij COEFFICIENTS CONTRASTS c 1 c 2 c ARE THEY ORTHOGONAL? SET 1 cijci’j ∑cijci’j COEFFICIENTS CONTRASTS c 1 c 2 c](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-42.jpg)
ARE THEY ORTHOGONAL? SET 1 cijci’j ∑cijci’j COEFFICIENTS CONTRASTS c 1 c 2 c 3 c 4 1 -1 - - Y 1 -Y 2 - - 1 -1 Y 3 -Y 4 ½ ½ -1/2 1/2 -1/2 ½ (Y 2+Y 2)/2 – (Y 3+Y 4)/2 0 YES!. . . BUT THEY DON’T HAVE TO BE
![DOING PLANNED ORTHOGONAL CONTRASTS USING SAS SAS INPUT dataset Give your dataset an DOING PLANNED ORTHOGONAL CONTRASTS USING SAS � SAS INPUT: dataset; Give your dataset an](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-43.jpg)
DOING PLANNED ORTHOGONAL CONTRASTS USING SAS � SAS INPUT: dataset; Give your dataset an informative name. input y a; Tell SAS what you’ve inputted: column 1 is cards; your y variable (dependent) and column 2 is 3 1 your a variable (independent). 4 2 7 3 This is followed by your actual data. 7 4. . . “Model” tells SAS that Use SAS procedure proc glm; class a; “Class” tells SAS that a is categorical. you want to look at the effects that a has on y. model y = a; contrast 'Compare 1 and 2' a 1 -1 0 0; Indicate “weights” (kind of contrast 'Compare 3 and 4' a 0 0 1 -1; contrast 'Compare 1 and 2 with 3 and 4' a 1 1 -1 -1; like the coefficients). run; Then enter your “contrast” statements.
![DOING PLANNED ORTHOGONAL CONTRASTS USING SAS ANOVA DOING PLANNED ORTHOGONAL CONTRASTS USING SAS ANOVA](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-44.jpg)
DOING PLANNED ORTHOGONAL CONTRASTS USING SAS ANOVA
![RECOMMENDED READING Kirk RE 1982 Experimental design procedures for the behavioural sciences Second ed RECOMMENDED READING �Kirk RE. 1982. Experimental design: procedures for the behavioural sciences. Second ed.](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-45.jpg)
RECOMMENDED READING �Kirk RE. 1982. Experimental design: procedures for the behavioural sciences. Second ed. CA: Wadsworth, Inc. �Field A, Miles J. 2010. Discovering Statistics Using SAS. London: SAGE Publications Ltd. �Institute for Digital Research and Education at UCLA: http: //www. ats. ucla. edu/stat/ Stata, SAS, SPSS and R.
![A PRIORI OR PLANNED CONTRASTS THE END A PRIORI OR PLANNED CONTRASTS THE END](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-46.jpg)
A PRIORI OR PLANNED CONTRASTS THE END
![EXTRA SLIDES EXTRA SLIDES](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-47.jpg)
EXTRA SLIDES
![PLANNED COMPARISONS USING A t STATISTIC t cj Yj MSerrorcjnj Where c is PLANNED COMPARISONS USING A t STATISTIC �t = ∑cj. Yj/ √MSerror∑cj/nj Where c is](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-48.jpg)
PLANNED COMPARISONS USING A t STATISTIC �t = ∑cj. Yj/ √MSerror∑cj/nj Where c is the coefficient, Y is the corresponding mean, n is the sample size, and MSerror is the Mean Square Error.
![PLANNED COMPARISONS USING A t STATISTIC For Example you want to compare 2 means PLANNED COMPARISONS USING A t STATISTIC �For Example, you want to compare 2 means:](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-49.jpg)
PLANNED COMPARISONS USING A t STATISTIC �For Example, you want to compare 2 means: Y 1 = 48. 7 and Y 2=43. 4 c 1 = 1 and c 2 = -1 n=10 MSerror=28. 8 �t = (1)48. 7 + (-1)43. 4 [√ 28. 8(12/10) + (-12/10)] = 2. 21
![CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS Y 1 Y 2 tα2 vSE ψ CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �(Y 1 -Y 2) – tα/2, v(SE) ≤ ψ](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-50.jpg)
CONFIDENCE INTERVALS FOR ORTHOGONAL CONTRASTS �(Y 1 -Y 2) – tα/2, v(SE) ≤ ψ ≤ (Y 1 -Y 2) – tα/2, v(SE) �For Example: If, Y 1 = 48. 7, Y 2 = 43. 4, and SE(tα(2), df) = 4. 8 (Y 1 -Y 2) – SE(tα(2), df) = 0. 5 (Y 1 -Y 2) + SE(tα(2), df) = 10. 1 Thus, you can be 95% confident that the true value of ψ is between 0. 5 and 10. 1. Because the confidence interval does not include 0, you can also reject the null hypothesis that Y 1 -Y 2 = 0. (Example after Kirk 1982)
![TYPE I ERRORS AND ORTHOGONAL CONTRASTS As the number of independent tests increases so TYPE I ERRORS AND ORTHOGONAL CONTRASTS �As the number of independent tests increases, so](https://slidetodoc.com/presentation_image/47161884b50aac71c8380e355c28fc7b/image-51.jpg)
TYPE I ERRORS AND ORTHOGONAL CONTRASTS �As the number of independent tests increases, so does the probability of committing a Type 1 error. �For Example, when α = 0. 05: 1 -(1 -0. 05)3=0. 14 (C=3) 1 -(1 -0. 05)5=0. 23 (C=5) 1 -(1 -0. 05)10=0. 40 (C=10) (Example after Kirk 1982)
Planned comparisons example
Planned contrasts spss
A wise economist asks a question analysis
Ace different tests iq tests but
Comparison test series
Multiple test adalah
Tukey multiple comparison method confidence interval
North carolina land of contrasts
Contrast and contradiction examples
Chapter 5 a land of contrasts answer key
Example of sarcasm
Contrasts and contradictions signpost
What does contrasts and contradictions mean
Multiple baseline vs multiple probe design
Multiple instruction multiple data
Syntetisk a priori
Niezgodność prakseologiczna
A priori and post hoc comparisons
Objects of knowledge
Maciej pichlak
Synthetic a priori judgments
Lex posterior derogat legi priori
Differenza tra certezza e verità
Segmentazione a priori
Conocimiento a posteriori ejemplos
Priori vs posteriori
A priori vs a posteriori
Ejemplos imperativo categorico
Lex posterior derogat legi priori
El idealismo segun kant
Noumeno kant
Metode a posteriori
Luka aksjologiczna
A priori knowledge
Cual es el origen del conocimiento
Maciej pichlak
Metodo deductivo ejemplos
Apriori machine learning
Giudizi sintetici a priori
Examples of priori codes
Systematyzacja pozioma
A priori kant
Planned giving certification
Planned cities on the indus worksheet answers
Types of planned maintenance
Grandfolkie planned change theory
System capacity definition in operations management
Critique of planned change
Doing business with fdot
Command economy pros
104-r planned academic worksheet
Tipe lokasi bisnis ritel