Survey research and design in psychology/Assessment/Exams/Final/Practice

This page addresses mistakes and questions raised by students about the 2009 practice final exam.

19
{ If the r between two IVs is 0, then R will equal the r of X1 and Y plus the r of X2 and Y. + True - False
 * type=""}

Am I right in thinking that r is correlations coefficient? if so, then 0 = no relationship? R is measure of effect size... I don't follow the rest of the question... Similar follows for Q20. ✅ Yes, r is the statistical abbreviation for Pearson's linear correlation. I'll add that to the list of abbreviations at the start of the exam for future. R is the multiple correlation coefficient (see Multiple linear regression). I suggest you try drawing a Venn diagram. If the r between the predictors is 0, then they are independent. Therefore, R will be the r between predictor 1 and the DV plus the r between predictor 2 and the DV. -- Jtneill - Talk 05:53, 13 June 2008 (UTC)

20
{If the r between two IVs is 1, then R will equal the r of X1 and Y plus the r of X2 and Y. + True - False
 * type=""}

✅ As per 19, however, if the predictors are perfectly correlated, then R will equal the r between one of the predictors (either one) and the DV. Again, sketching a Venn Diagram is a good idea. -- Jtneill - Talk 05:53, 13 June 2008 (UTC)

30
{A MLR was conducted with Attentiveness in Year eight as the DV and three IVs. Here is the regression coefficients table.

Which of the following statements is false? + These results are unreliable, since some of the assumptions of MLR have been violated. - Attentiveness and sociability have a significant positive linear r. - Sociability is about twice as important as a predictor compared to settledness. - Gender doesn't play much of a role in the overall results for this multiple linear regression. (n other words, if gender was removed from the analysis, the main conclusions would be similar.
 * type=""}

How can you tell from this table? ✅ This question could be answered by the process of elimination: b, c, d and true statements because:
 * b: There is a notable drop between the zero-order and partial correlations for attentiveness and sociability (but not so much for Sex of student). Therefore these predictors must be correlated with one another.
 * c: The Beta for for Sociability is about twice as large as for Settledness.
 * d: Sex of student is not significant and has a fairly small Beta, so its not adding much in the way of predicting variance in the DV.
 * -- Jtneill - Talk 06:42, 13 June 2008 (UTC)

35
{Here is a famous rule of thumb for assessing computed t values: We can instantly treat an associated regression coefficient as statistically significant at the 95% level of confidence whenever: - The magnitude of a computed t value is negative. - The magnitude of a computed t value is positive. + The absolute magnitude of a computed t value exceeds 2. - The absolute magnitude of a computed t value exceeds 5. - None of the above.
 * type=""}

Which lecture did you go through this one, so I could go over it? ✅ I suggest reading up on the properties of the normal distribution which were discussed in the ANOVA lecture and reviewing t-tests, which is assumed knowledge. -- Jtneill - Talk 06:41, 13 June 2008 (UTC)

45
In MLR, if the N was halved (from 200 to 100), Adjusted R2 will: __________.


 * a. not change
 * b. be doubled
 * c. be halved
 * d. increase slightly
 * e. decrease slightly

How do you work this out? ✅ I suggest you read about Adjusted R2. Basically, this reduces R2 (for a sample) in order to estimate variance explained in a population. The larger the sample size, the smaller the reduction for adjusted R2. -- Jtneill - Talk 06:46, 13 June 2008 (UTC)

80-83
For a 2 (A-B) x 2 (-/Y) factorial ANOVA, the following cell means indicate what kind of effects? (insert various examples of means displayed in a 2 x 2 table)
 * a. Main effect for X-Y, no main effect A-B, no interaction
 * b. Main effect for A-B, main effect for X-Y, no interaction
 * c. Main effect for A-B, main effect for X-Y, interaction
 * d. Main effect for A-B, no main effect X-Y, interaction
 * e. Main effect for X-Y, no main effect A-B, interaction

How can you work that out? I am unable to understand how I interpret this table in regards to the questions. I have looked through the textbook and lecture notes and I can't seem to find it. Could you please point me in the right direction? ✅ I suggest sketching 2 x 2 line graphs with the Y axis indicating mean values for either:
 * (i) X and Y on the x-axis and separate lines for A and B; or
 * (ii) A and B on the x-axis and separate lines for X and Y.

For the 2 x 2 table, it can also be helpful to work out the margin totals, i.e, the average scores for X, Y, A and B. Compare X and Y scores for one main effect. Compare A and B scores for the other main effect. And check whether or not the lines in the graph are parallel for the interaction effect. I understand by drawing a line graph that you can tell if there is an interaction effect but how do you know if there are main effects which are significant. Do you just assume that if the numbers are different that there is an effect? ✅Yes, assume a sufficiently powerful study which would detect as significant any "obvious" differences between the means of the cells and/or groups.

84
What is Cohen's rule of thumb for interpreting $$\eta$$2?


 * a. .01 small, .03 medium, .1 large
 * b. .1 small, .3 medium, .6 large
 * c. .01 small, .06 medium, .14 large
 * d. .1 small, .3 medium, .5 large

Where is this guideline from? ✅ Oops, sorry, this wasn't in the unit materials, that's a mistake; Cohen's (1988) guideline which is: 0.01 small, 0.059 medium, 0.135 large
 * . -- Jtneill - Talk 07:04, 13 June 2008 (UTC)

87
Consider an experiment with 2 factors, A and B, and a response, Y. Which of the following are true:


 * i. The interaction term is significant when the change in the true average response Y when factor A changes is the same for each level of B.
 * ii. The interaction term is significant when the main effects provide an incomplete description of the data.
 * iii. The interaction term is significant when the effect of A on the true average response depends on what level of factor B is considered.
 * iv. If we create an interaction plot and the lines are not parallel, we can conclude that there is sufficient statistical evidence for an interaction effect.


 * a. i & ii
 * b. i, ii, & iii
 * c. ii & iii
 * d. iii & iv
 * e. ii, iii, & iv

Can you explain how this works? Im not sure about ii? ✅ ii. When there is an interaction, then the main effects do not sufficiently explain the relations between the IVs and DV. Thus, the interaction explains additional variance in the DV above and beyond the main effects. When the interaction is not significant (or large) then the main effects provide a sufficient explanation and can be interpreted on their own. -- Jtneill - Talk 07:16, 13 June 2008 (UTC)

104
104. In a convenience sample survey, which aspects of the study is the researcher likely to have most control? (choose least to most control)


 * i. N
 * ii. ES
 * iii. Critical $$\alpha$$


 * a. Critical $$\alpha$$, N, ES
 * b. N, ES, Critical $$\alpha$$
 * c. Critical $$\alpha$$, ES, i
 * d. ES, N, Critical $$\alpha$$
 * e. N, ES, Critical $$\alpha$$

I thought that controlling sample size, even in convenience sample, was the easiest part? ✅ This is going to vary from somewhat study to study, and is somewhat subjective. Often N is limited by external contraints (e.g., time/money), but of course a researcher does ultimately usually decide and can control the sample size. ES one may have control over. e.g, in an experimental study, the "dosage" can be controlled, but in survey-type studies one tends to have less control. Critical alpha is very controllable. -- -- Jtneill - Talk - c 14:11, 27 May 2009 (UTC)

FAQ
Will we need to know the formulas?

✅ You aren't expected to know or memorise formulae per se, but understanding formulae will be beneficial for understanding how key statistics are derived, what they are used for, how to interpret, assumptions, etc.