Term
| If a null hypothesis is true, but you reject it, you have made what type of error? |
|
Definition
|
|
Term
| If a null hypothesis is not true, and you do not reject it, then you have made what type of error? |
|
Definition
|
|
Term
| As alpha goes up, beta goes ? |
|
Definition
|
|
Term
| As beta goes up, power goes ? |
|
Definition
| down; Remember that power is 1 - beta |
|
|
Term
| As alpha goes up, power goes ? |
|
Definition
|
|
Term
Choose: anova F-tests are always/sometimes one sided/two sided for determining a p-value. |
|
Definition
|
|
Term
What are the 5 assumptions for calculating ANOVA? |
|
Definition
Random samples-are required to be unbiased Indepedent samples If they're dependent (or repeated), you have to use repeated measures ANOVA. Independent observations w/in groups Normal distribution - (aka normal population) Equal variances (homogeneity) |
|
|
Term
| When calculating ANOVA, if you have unequal variances, you have to the data. Name three ways that you would do this to your data. |
|
Definition
transform; log, natural log, square root |
|
|
Term
is the sum of squares (SS) What is SSB? What are other names for SSB? What is SSW? What are other names for SSW? Choose: The signal/noise should be > than the signal/noise, otherwise, we won't be able to detect the signal/noise. |
|
Definition
Variance; It is the variance b/w groups (it is the difference of each mean from the overall mean); SS(treatment) or SS(effects) or signal; It is the variance w/in groups; SS(error) or noise. signal-noise-signal |
|
|
Term
|
Definition
| It is the mean square b/w groups |
|
|
Term
SSB = 43.21 SSW = 50.68 T or F: the error is greater than the treatment Why or why not? |
|
Definition
False, b/c you still have to divide by the df. The SSB will have a much smaller df (eg, 2) than the SSW (eg, 39). After dividing the SSB and SSW by their respective df's, you get MSB = 21.61 and MSW = 1.30. |
|
|
Term
| If a comparison is not planned ahead of time, you must use even if only one post-hoc comparison was performed. However, you usually perform all comparisons. |
|
Definition
|
|
Term
If comparisons are selected a priori, what does c =? If comparisons are selected post hoc, what does c =? |
|
Definition
a priori: c = number of comparisons to be made post hoc: c = total number of possible comparisons |
|
|
Term
What is the point of the Bonferroni method? |
|
Definition
| It's like an adjusted alpha, labeled alpha'. It increases the CL for each CI in order to ensure a specific overall CL. Table 11.4 shows how conservative the bonferroni method is. |
|
|
Term
When should you use bonferroni? When should you use tukey? |
|
Definition
The bonferroni method should be used when comparisons are selected a priori. The Tukey method should be used when ALL comparisons are selected - it doesn't matter if they're selected a priori or post hoc. If comparisons are selected post hoc, use the Tukey method. |
|
|
Term
| The same three factors that affect statistical outcomes affect ANOVA results. What are they? |
|
Definition
Treatment effects (differences b/w means) Variance (SSB compared to SSW) # of subjects |
|
|
Term
| What are the 4 assumptions of the regression model? |
|
Definition
NICL Normal distribution Independence (y observations are independent of each other) Constancy of variance (the sd of y is constant for each x) Linear relationship (as opposed to exponential, curvilinear, etc) |
|
|
Term
What is residual analysis used to test? |
|
Definition
| The assumptions of linear regression. |
|
|
Term
Correlation analysis uses pearson, r. What are the 3 assumptions of pearson? |
|
Definition
1) Both the ind & dep variables must b continuous. 2) Dependent variable must be rational 3)No multicollinearity (meaning ind variables cant be related to other ind variables) |
|
|
Term
| What are the assumptions of B? |
|
Definition
1) Each y observation is independent 2) All potential y values are normally distributed for each x value 3) The std deviation of y is constant for each x 4) The sampling distribution of b is normal |
|
|
Term
| What are the 4 assumptions for r? |
|
Definition
1) Each y observation is independent 2) The sampling distribution of both x and y are normal. 3) The std deviation of y is constant for each x 4) The sampling distribution of r is normal. |
|
|
Term
If all population means are , the sample means will be very close to each other and close to the . The weighted square differences will be , and a F will be calculated. |
|
Definition
| equal; grand mean; small; small |
|
|
Term
| ANOVA T or F: Variance cannot be negative |
|
Definition
| True. If it is, there is too much error. |
|
|
Term
| If some population means are quite different, corresponding sample means will be than each other and the . Corresponding weighted squares will be , the and will be small, and the F will be . |
|
Definition
| quite different; grand mean; large; SSB; MSB; large |
|
|
Term
| As alpha increases, Fcrit . |
|
Definition
|
|
Term
| When is one-way ANOVA used? |
|
Definition
| When you need to compare more than two means. More specifically, when you have one factor w/ more than two levels of the factor (ie, Drug A, Drug B, and DrugC). |
|
|
Term
|
Definition
Error slippage occurs when you try to do a bunch of t-tests when you really should have done ANOVA. You'll think you have significant differences, but you really don't, b/c the alpha is slipping from 0.5 to 1. |
|
|
Term
|
Definition
| It means group 3, observation 2. |
|
|
Term
| If you had two or more independent variables, what type of ANOVA would you use? |
|
Definition
| Factorial ANOVA. The different factors may be Drugs A, B, & C (1st factor), Age (second factor), & gender (third factor). |
|
|
Term
| When would you use one-way ANOVA? Name a requirement of ANOVA. |
|
Definition
| Use one-way ANOVA when you have one factor and more than two levels. The samples in the different levels must be independent of each other. |
|
|
Term
| When do you use post-hoc analysis? |
|
Definition
| When the null has been rejected and you're trying to find out which mean(s) are different. |
|
|
Term
| When do you use repeated measures ANOVA? |
|
Definition
| If measures are dependent (or repeated), you use a crossover design called repeated measures ANOVA. |
|
|
Term
What is the proper term used when your putting variances into groups? What is another name for variance? |
|
Definition
Partitioning Sums of Squares |
|
|
Term
| SSB & SSW are weighted by the . |
|
Definition
|
|
Term
| Choose: When p-Value is greater/lesser than or equal to alpha, H0 cannot be rejected. |
|
Definition
|
|
Term
| Which is worse: a type I or type II error? |
|
Definition
|
|
Term
A type I error is aka A type II error is aka |
|
Definition
|
|
Term
Define alpha What does it mean when we set out alpha to 0.05? |
|
Definition
It is the probability that a difference does not exist when the study shows that a difference does exist. If a study was repeated 100 times, we would be willing to be wrong 5 times. We would conclude 5 times that a difference exists when in fact it does not. |
|
|
Term
T or F: A rejection region will be larger for a one-tail alpha than a two-tail alpha |
|
Definition
|
|
Term
|
Definition
| It is the percent chance that a difference b/w experimental grps does exist even though the study concludes no difference. |
|
|
Term
T or F: We usually look at beta after an experiment, to find out why we DIDN'T reject the null. Beta is usually set at the beginning of an experiment. Where should it be set, and why do we set it? |
|
Definition
True Beta should be less than or equal to 0.2. We set beta, b/c we want a certain power (at least 80% or 0.80). Notice that this isn't as strict as alpha. Remember, that we are more willing to be wrong on a type II than a type I error. |
|
|
Term
What are 3 reasons for a type II error? Which is the most common? |
|
Definition
Better response rate in the comparator grp Increased variance Too few subjects (most common) |
|
|
Term
| If type I & type II error are inversely related, how can we reduce the chance of a type I error w/o increasing the type II error? |
|
Definition
|
|
Term
What is power? How do you calculate power? |
|
Definition
It is the percent chance that if a difference b/w grps exists that your study will detect it statistically. power = 1 - beta |
|
|
Term
Name the 3 factors that affect power |
|
Definition
Size of treatment effect, alpha, and variability. |
|
|
Term
| The size of treatment is aka |
|
Definition
|
|
Term
Power is affected by the "effect size". If there is a large effect, the difference is easy/difficult to detect and power is high/low. |
|
Definition
|
|
Term
Power is affected by the effect size. If there is a small effect, the difference will be easy/difficult to detect, type II error will be higher/lower, and power will be higher/lower. |
|
Definition
|
|
Term
| A larger variance will result in a lower/higher power |
|
Definition
|
|
Term
Why is it that an increased sample size increases power? |
|
Definition
| Because it increases df, and it controls variance (ie, SEM goes down, b/c SEM = s / sq root of n) |
|
|
Term
| Why should power (and beta) be calculated a priori ? |
|
Definition
| To determine the sample size needed for a study. |
|
|
Term
| When does power not have to be calculated? |
|
Definition
| When there is a significant difference. |
|
|
Term
T or F: Statistical significance does not equal clinical significance |
|
Definition
|
|
Term
When do you have to do post-hoc analysis? |
|
Definition
If the null is rejected, you have to do post-hoc analysis. Otoh, if the null accepted, the analysis is complete, b/c the variances are equal. |
|
|
Term
| What is linear regression? |
|
Definition
| To test for associations b/w variables. If independent values are given, we can use linear regression to predict the dependent variable. |
|
|
Term
| Why is the simple regression equation different from the sample regression equation? |
|
Definition
| B/c simple regression deals w/ population, which means error is included in the equation. Error is not included in the sample linear regression equation. |
|
|
Term
| Least squares regression is often used to calculate linear regression. What is the least squares regression line used for? |
|
Definition
it sums the squared distance of the actual values to make them as close to what is predicted as possible. |
|
|
Term
If you see the least squares line equation, y = 560 + 0.14X, what does it mean? |
|
Definition
| A 0.14 change in x will bring about a 1 unit change in Y. |
|
|
Term
|
Definition
| It's the difference between the actual y-value and the estimated y-value. |
|
|
Term
| What is a residual analysis used for? |
|
Definition
| To test the assumptions of linear regression. It basically diagnoses problems w/ a model. |
|
|
Term
| If you got a curve shaped residual scatter plot, would this be good or bad? Why? |
|
Definition
| This would be bad. It would reveal that our data is not linear. We would have to transform the data. |
|
|
Term
| What are the 2 problems w/ the funnel shaped residual scatter plot? |
|
Definition
1) The funnel reveals a problem w/ independence. 2) There is a constancy of variance. We have heteroscedasticity when we want homoscedasticity. |
|
|
Term
Why is the coefficient of determination important? The closer r2 is to , the better the regression model. |
|
Definition
It reveals how much variation is explained by the regression line. |
|
|
Term
| What is the difference b/w the coefficient of determination and the correlation coefficient? |
|
Definition
| The coefficient of determination is the variance accounted for by a whole model. It includes the y-variables, whereas the correlation coefficient doesn't. The correlation coefficient lets you know if 2 variables are associated. The correlation coefficient b/w dependent and independent variables will generally be less than the coefficient of determination, b/c the model, which has y-variables will account for more variance. |
|
|
Term
| What is multiple linear regression and what makes it different than correlation? |
|
Definition
| It's used when you want to fit multiple independent variables to one dependent variable. Fe, the dependent variable is a students fall semester GPA. The independent variables could be GPA, ACT, ethnicity, gender, age, etc. It's different than a correlation, b/c it has predictive abilities. Fe, it can predict a dependent variable from one or more independent variables. |
|
|
Term
Statistical significance is best evaluated w/ . Clinical significance is best evaluated w/ . |
|
Definition
correlation coefficient (r) coefficient of determination (r2) |
|
|
