psych 350 exam 2 essays
Describe effect size estimates, tell how they are related to significance tests, the information they provide that is not provided by significance tests.
Effect size estimate, r, is a one number quantitative summary about how strong is relationship between variables. The sign is concerned with the type of relationship while the size is concerned with the magnitude of the relationship. Significance tests only search for whether there is a significant pattern or not while effect size estimates tell the magnitude of this pattern. The larger the effect size, the more likely you are to reject to null.
Describe the process of NHST, tell the (five) possible outcomes and tell the likely reasons for each. (Be sure to tell what this acronym means.)
Null hypothesis significance testing is what NHST stands for. First you must come up with a research hypothesis. You will want to determine what the null hypothesis is as well. The research hypothesis can be the null hypothesis. You will then collect data from a representative sample of the population. You will then run a data analysis and get the significance statistic and p-value. You decide to reject or retain the null hypothesis based on the p-value. If p is less than .05 then you reject the null. If p is more than .05 you retain the null and reject your research hypothesis. You can 1.correctly retain the null hypothesis. 2. correctly reject the null hypothesis. With either of these (#1 or 2) It is likely that you had good measures, a representative sample, and a good design in your study. You could also 3. commit a type I error - false alarm, you found an effect when really there is not an effect in the population. 4. a type II error- miss - you did not find an effect but really there is an effect in the population. or 5. a type III error - misspecification - you found an effect and there is an effect in the population but you found a false one. Type I and III errors can be a result of a bad design, a nonrepresentative sample, and bad measures. A type II error is likely to result from a bad design, nonrepresentative sample, bad measures or a sample size that is too small.
Compare and contrasts the "interesting pairs" of the four bivariate data analysis models we are working with.
One "interesting pair" is WG ANOVA and BG ANOVA; both of these analyses are ANOVAs and so both are used to compare one qualitative and one quantitative variable. They are different, because WG ANOVA compares data from a within-groups experiment, whereas, BG ANOVA compares data from a between groups experiment. Another pair is BG ANOVA and Pearson's Chi-square; both of these analyses are generally used for between groups experiments. They are different, because BG ANOVA is used to compare one quantitative variable and one qualitative variable, whereas, Pearson's Chi-square is used to compare two qualitative variables. A third pair is WG ANOVA and Pearson's correlation; both of these analyses are generally used for within-groups experiments. They are different, because WG ANOVA is used to compare one quantitative variable and one qualitative variable, whereas, Pearson's correlation is used to compare two quantitative variables. The fourth and final pair is Pearson's Chi-square and Pearson's correlation; both of these analyses are used to compare two of the same type of variables. They are different, because Pearson's Chi-square is used to compare two qualitative variables, whereas, Pearson's correlation is used to compare two quantitative variables.
Tell when to use a Pearson's correlation, the possible research hypotheses for this statistical model, and when correlation can be used to test each type of Research Hypothesis (attributive, associative and causal).
Pearson's correlation is utilized when you have two quantitative variables and you wish to see if there is a linear relationship between those variables. Your research hypothesis would represent that by stating that one score affects the other in a certain way. The correlation is affected by the size and sign of the r. A small size makes for a small effect. A positive correlation results in a positive linear relationship where as a negative r results in a negative linear relationship. There could also be no relationship between the variables. This statistic should only be utilized after consulting a scatterplot and seeing there is a possibility of having a linear relationship. This bivariate statistic cannot be utilized for attributive hypothesis because it involves the interaction of two variables, not the distinction of a single variable. Associative hypothesis can be tested at all times because they can be non-experiments or true experiments. Causal hypothesis can only use correlational data if the hypothesis is a true experiment.
What is meant by "statistical power" and what is the advantage if our research has lots of it? Describe how power analyses are conducted and how they can inform our statistical decisions.
Statistical power is the power you have to reject the null hypothesis. It is an advantage to have lots of power in a study because it decreases the probability that you made a type II error. Power analyses can be done two different ways. Before you begin the study you can decide how much of a risk you are willing to take on making a type II error and find how many subjects you will need in your study to not have a type II error via an a priori power analysis. After a study is complete and you maintained the null, you can conduct a post hoc power analysis to find out the probability that you made a type II error, and how many subjects you will need to conduct a replication of your study in order to not make (possibly another) type II error.
Tell when to use each type of ANOVA, the possible research hypotheses for this statistical model, and when ANOVA can be used to test each type of Research Hypothesis (attributive, associative and causal).
There are two types of ANOVA, between groups ANOVA and within groups ANOVA. Between groups ANOVA is used with a between groups design, one in which participants are in different groups and these groups are subject to different conditions for the study (ex, cognitive behavioral therapy for depression and folksinging groups to aid depression). A within groups ANOVA is used with a within groups design, in which all participants are subject to all research conditions (ex, depression test scores taken before and after therapy). There are three possible research hypotheses for these two group models. It can be hypothesized that one group will do better than the other, worse than the other, or that both groups will do the same.ANOVA can never be used to test an attributive hypothesis because an attributive hypothesis is a single-variable hypothesis. ANOVA can be used to test an associative hypothesis at any time, regardless of the research design. In testing a causal hypothesis, ANOVA can only be used if there is a true experiment with random assignment of individuals, manipulation of the IV, and no confounds.
Respond to and describe the statement, "Rejecting the null hypothesis guarantees support for the research hypothesis."
This statement is false. Rejecting the null hypothesis means that there is (according to the data) a relationship between the two variables in question in the population. There are two reasons a rejection of the null does not guarantee support for the research hypothesis. One reason is that the research hypothesis might have been the null, researchers are allowed to choose this as their hypothesis. If this were the case, then rejecting the null hypothesis actually provides no support for the research hypothesis. Another reason is that the data may be significant, and the null rejected, but the observed relationship may be different from the hypothesized one. For instance, you could expect a negative linear relationship between number of times brushing teeth and number of cavities but instead find a significant positive linear relationship (perhaps the children are brushing with sugar, not toothpaste?). Rejecting the null guarantees (assuming it is a correct rejection) that a relationship exists and says nothing for supporting the research hypothesis outright.
Tell when to use Pearson's X², the possible research hypotheses for this statistical model, and when X² can be used to test each type of Research Hypothesis (attributive, associative and causal).
We use Pearson's X^2 when there is two qualitative variables. The research hypothesis will talk about a pattern of relationship between the two variables. The possible research hypothesis for this model is there is an asymmetric pattern of relationship, there is a symmetric pattern of relationship between the two variables, and there is no pattern of relationship between the two variables. Pearson's X^2 can be used to test any associative research hypothesis, and can be used to study a causal research hypothesis providing that the causal study has no confounds, random assignment of individuals, and everything needed to make a good study. It cannot be used to test attributive hypotheses because those are univariate.
