MMB-101-questions Flashcards
- If more than one a priori contrast is performed, the significance level (alpha) that is used must be multiplied by the number of comparisons, to prevent inflation of type I error rates.
True/ False
FALSE
The significance level should be divided by the number of comparisons, if using a Bonferroni correction. The test needs to be tougher, not laxer.
- In repeated measures designs, carry over effects (fatigue, etc) are controlled for by presenting the experimental conditions in the same order for every participant.
True/ False
FALSE
- In a line plot of two IVs, an interaction will show itself as a lack of parallelism between the lines.
True/ False
TRUE
- Type I error refers to the probability of falsely rejecting the null when it is true.
True/ False
TRUE
- Cohen’s “d” is defined as the population standard deviation divided by the difference between the group means.
True/ False
FALSE
Cohen’s d can be calculated as the difference between the means divided by the pooled SD
M1-M2/SD
Cohen’s d is an effect size used to indicate the standardised difference … the size of difference between each of the pairwise mean differences.
- An application for ethical approval for an experiment should be made in advance of any design, planning or other preparations for the experiment.
True/ False
FALSE
- Testing the null hypothesis with correlations involves determining if the observed sample correlation r is significantly different from zero.
True/ False
TRUE
- The Bonferroni correction for multiple testing is applied by using a stricter alpha value, found by dividing some conventional value (such as .05) by the number of comparisons.
True/ False
TRUE
- The ‘imputation’ of values for missing data is recommended whenever there is a systematic bias in the pattern of missing data points.
True/ False
FALSE
- When using multifactorial ANOVA, you should never combine repeated measures and independent groups IVs in the same analysis.
True/ False
FALSE
There is no reason for not combining two types of variables in a “mixed” ANOVA.
The correct answer is ‘False’.
- A common assumption of parametric tests involving group comparisons on some DV is that the distribution of the DV averaged over all the groups is normal.
true/false
FALSE
The assumption is that the distribution within groups is normal. If you take the whole dataset it may well not be normal – if there is a large effect size very often it will not be normal, even if the groups themselves are.
- High reliability implies small random error; high validity means small systematic error.
True/ False
TRUE
- For a test with a symmetric distribution and for a given set of data, you can find the p value for a one-tailed test of these data by halving the p value for the corresponding two tailed test.
True/ False
TRUE
This is why one-tailed tests are (if appropriate) more powerful than two-tailed tests.
The correct answer is ‘True’.
- When you decide to conduct a two-tailed test rather than a one-tailed test, at the same level of alpha, power increases.
True/ False
FALSE
- As a post hoc test in ANOVA, the Games Howell test tends to be less powerful than the Tukey HSD test where the Tukey test can be applied, but it does not require the assumption of homogeneity of variance so can sometimes be used where the Tukey test cannot.
True/ False
TRUE
Another example of the trade-off between power and robustness (ie Games-Howell is more robust than Tukey’s HSD).
The correct answer is ‘True’.
- Exploratory factor analysis basically involves identifying clusters of variables that are correlated with other variables within the cluster, but not strongly correlated with variables in other clusters.
True/ False
TRUE
- In estimating the number of factors from a scree plot in exploratory factor analysis, one should include the “elbow” where the break in slope occurs.
True/ False
FALSE
You should not include the elbow: it is part of the “noise” variance represented by the scree.
- The p value yielded by a statistical test in SPSS is the probability of getting a result at least as extreme as that observed, if the null hypothesis is false.
True/ False
FALSE
It is the probability of getting a result at least as extreme as that observed, if the null hypothesis is true.
- Fisher’s null hypothesis significance tests enable a null hypothesis to be accepted if an event happens which is unlikely on the assumption of the hypothesis.
True/ False
FALSE
Yes, the statement is false. An NHST enables the null hypothesis to be rejected if a sufficiently unlikely event occurs. NHSTs never allow the null to be accepted or confirmed.
The correct answer is ‘False’.
- The null hypothesis in a test of association between two categorical variables, each taking two values, is that each member of the population occurs once and only once in each of the four “cells” comprising the two variables.
True/ False
FALSE
This is one of the assumptions of the chi-square test. The null hypothesis is that there is no association between the variables in the population.
The correct answer is ‘False’.
- Fisher’s F test looks at the ratio between the measures of variability between, and within groups, and declares a significant result if this ratio is small enough.
True/ False
FALSE
The F-ratio is a ratio of signal to noise, so it needs to be large enough, for significance.
The correct answer is ‘False’.
- If the null hypothesis is true, the group means for the data from a sample will all be exactly equal.
True/ False
FALSE
Group means will almost inevitably differ in a sample, due to the random error or noise element involved in sampling from a population, even if population means are equal.
The correct answer is ‘False’.
- The gold standard for providing evidence of causality is the doubleblind, randomised controlled trial.
True/ False
TRUE
- Multiple regression requires a parametric DV and parametric and/or dichotomous (twovalued) IVs (categorical IVs with more than two levels must be dummy coded).
True/ False
TRUE
It also of course requires a lot of other assumptions on lack of outliers, normal distribution of errors etc.
The correct answer is ‘True’.
- The following set of contrast coefficients tests whether level 1 of a categorical independent variable has a different mean from the average of levels 2 and 3:
(0, 0.5, 0.5).
True/ False
FALSE
What these coefficients in fact do is to compare levels 2 and 3 with one another, ignoring level 1.
The correct answer is ‘False’.
- At the rotation stage of an Exploratory Factor Analysis, it is a good idea to use an orthogonal rotation such as varimax in the first instance, unless you have good reason to suppose that factors are oblique.
True/ False
FALSE
It is generally better to use an oblique method unless you have good reason to suppose the factors are orthogonal. This is because imposing an orthogonal rotation on factors which are in fact oblique may give misleading results, whereas an oblique rotation will reveal if factors are actually (or nearly) orthogonal.
The correct answer is ‘False’.
- With an oblique rotation, you should examine the structure matrix rather than the pattern matrix for evidence of a simple factor structure.
True/ False
FALSE
- The value under the R Square column in a multiple regression output refers to the proportion of variance in the dependent variable accounted for by the independent variables collectively, using a regression model.
True/ False
TRUE
Sometimes it’s better to use adjusted-R squared for this, but essentially the statement is true.
The correct answer is ‘True’.
- The presence of an interaction between two factors means that the effect of one factor on the DV varies, as the level of the other factor alters.
True/ False
TRUE
- A high value for Cronbach’s alpha for a set of test items always demonstrates a satisfactory outcome, and makes it unnecessary to check separately whether the instrument is truly unifactorial.
True/ False
In Rory’s view: Cronbach’s alpha is a rather crude test of internal consistency and should generally combined with checks for multifactoriality.
The correct answer is ‘False’.
31.
2 PART Q 1-30
- Possible alternatives to Fisher’s F test that are often used are a priori contrasts, and post hoc comparison of means.
True/ False
TRUE
And in Rory’s view, these alternatives are greatly superior.
The correct answer is ‘True’.
32.
2 PART Q 130
- A serious failure of the statistical assumptions on which a parametric test is based, may require the use of a nonparametric alternative where this is available.
True/ False
TRUE
Sometimes it’s possible to transform the DV to satisfy the assumptions for a parametric test, but in that case one can argue that the assumptions are in fact satisfied, after some preliminary preparation.
The correct answer is ‘True’.
33.
- An “operational definition” of a variable or construct consists of specifying the set of activities required to measure that variable or construct.
True/ False
TRUE
Correct, and a fundamental concept in experimental design.
The correct answer is ‘True’.
34.
- A common null hypothesis when comparing a dependent variable between two groups is that the mean of the population of scores from which the first group was drawn is equal to the mean of the population of scores from which the second group was drawn.
True/ False
TRUE
35.
- As a design, repeated measures ANOVA tends to be less powerful than independent groups ANOVA.
True/ False
FALSE
36.
- Popper’s view is that science advances by testing and confirming hypotheses.
True/ False
FALSE
Popper thought it advances by testing and refuting hypotheses – this leads to formulating new and better hypotheses in a continual process of improvement.
37.
- Normality can be tested using the KolmogorovSmirnov
test, and should be nonsignificant.
True/ False
TRUE
Like most tests of assumptions, this should not be significant.
The correct answer is ‘True’.
38.
- Estimations of correlation are sensitive to the presence of outliers and restriction of range.
True/ False
TRUE
39.
- A Likert scale produces a variable on a ratio scale.
True/ False
FALSE
40.
- Values of “d” of 0.2, 0.5 and 0.8 are described as “small”, “medium” and “large” effect
sizes.
True/ False
TRUE
Correct: these are the rough classifications, for what they are worth.
The correct answer is ‘True’.
41.
- High reliability implies small systematic error; high validity means large random error.
True/ False
FALSE
42.
- It is necessary to make some estimate of likely effect size in advance, when doing the power calculation for a planned experiment.
TRUE
Yes. There is no way to calculate power without having at least a working assumption about effect sizes.
The correct answer is ‘True’.
43.
- The z value of a data point is its distance from the mean, measured in units of the standard deviation.
True/ False
TRUE
44.
- The null hypothesis when testing a model in multiple regression is that in the population, the DV correlates with some linear function of the IVs.
True/ False
FALSE
45.
- The analysis of contingency tables using the chi square test requires that each participant should contribute to one and only one cell of the table.
True/ False
TRUE
46.
- When planning a research project, an important early step is to specify a research question or hypothesis to be tested.
True/ False
TRUE
Correct, and not only important but, Rory would argue, essential.
The correct answer is ‘True’.
47.
- In a graphical display of dependent variable data in the form of a bar chart in SPSS, the bars will touch neighbouring bars.
True/ False
FALSE
A small point, but bars do not touch other bars. Only in a histogram does this happen.
The correct answer is ‘False’.
48.
- In standard repeated measures designs, each participant experiences all levels of the treatment variable.
True/ False
TRUE
49.
- ANOVA requires Levene’s test for homogeneity of variance to be significant.
True/ False
FALSE
ANOVA requires Levene’s test for homogeneity of variance to be non-significant.
The correct answer is ‘False’.
50.
- A basic assumption in many applications of repeated measures ANOVA is sphericity.
True/ False
TRUE
51.,
- Nonparametric tests are usually less powerful than their parametric equivalents.
True/ False
TRUE
52.
- An essential feature of science is that it makes statements which are testable.
True/ False
TRUE
53.
- The usual measure of dispersion in a population is the range.
True/ False
FALSE
54.
- The Type I error rate is always equal to the significance value (alpha) adopted for the test.
TRUE
55.
- The significance of an association between two continuous variables is often measured by Pearson’s chi square test.
True/ False
FALSE
The chi-square test is used for a pair of categorical variables.
The correct answer is ‘False’.
56.
- When sample size increases and the alpha level goes up, power increases.
True/ False
TRUE
57.
- “Adjusted R squared” in multiple regression is a measure of the amount of variance in the DV accounted for by the IVs collectively.
True/ False
TRUE
58.
- With a one factor ANOVA, the null hypothesis is that the means of the DV are equal for all the subpopulations from which the groups are drawn.
True/ False
TRUE
59.
- The outcome of a NHST will tell you whether an effect is present but not, directly, how much of an effect is present.
True/ False
TRUE
Correct. The mere fact of knowing a p-value will not tell you the effect size without further information, such as sample sizes.
The correct answer is ‘True’.
60.
- If the treatment and control means for some DV in a population are 40 units and 30 units, and the standard deviation for the DV is 20 units, then Cohen’s “d” is calculated to be 0.5.
True/ False
TRUE
Correct: the difference in means is 10 units, divided by the s.d. of 20 gives 0.5.
The correct answer is ‘True’.
61.
3 PART Q1-41
- When alpha (significance level) increases and effect size increases, power increases.
True/ False
62
3 PART Q1-41
- High reliability means small systematic error; high validity means small random error.
True/ False
FALSE
High reliability means small random error.; high validity means small systematic error.
63.
- The value of the unstandardised coefficient for a given IV in multiple regression is the amount by which the DV is predicted to change when that IV is increased by 1, the other IVs being held constant.
True/ False
TRUE
This is true, and it’s a basic fact which you need to memorize at the very least, and understand, better still.
64.
- With a significance (alpha) level of 5%, the power will be 95%.
True/ False
FALSE
No: there is no such simple link between alpha level and power. It is beta level (type II error rate) that is related to power in this simple manner.
65.
- An IQ of 70 corresponds to a z value of 2, if the mean is 100 and the s.d. is 15.
True/ False
TRUE
66.
- A correlation of 0.6 between two variables means the variables share 60% of their variance with one another.
True/ False
FALSE
No: shared variance is calculated as the square of the Pearson’s correlation, in this case 36%.
67.
- When designing an experiment, if it is required to calculate power, it is not normally necessary to make some prior estimate of effect size.
True/ False
FALSE
68.
- The definition of a “contrast”, is that it is a formula involving group means in which all the coefficients add up to one.
True/ False
FALSE
No: the coefficients add to zero.
69.
- An effect size is another measure of statistical significance; it can be defined only for a specific set of sample data, never for a population.
True/ False
FALSE
No: it is essentially a parameter of the population.
70.
- Approximately 95% of the points for a normally distributed population lie between z = –2 and z = +2.
True/ False
TRUE
71.
- If two variables are “independent”, this implies that their population correlation is zero.
True/ False
TRUE
This is in fact true. Independent variables will always have zero correlation, though two variables with zero correlations will not necessarily be independent.
72.
- A zvalue of 1 corresponds to the 50% percentile point in a normal distribution.
True/ False
FALSE
73.
- To avoid problems with multicollinearity in multiple regression, the value of tolerance should be below 0.25.
True/ False
FALSE
No: tolerance should be above 0.25 under the usual criterion.
74.
- If a pair of continuous variables are “orthogonal”, this means their population correlation is zero.
True/ False
TRUE
Yes, and in fact the two terms really are equivalent here.
75.
- Tukey’s “honestly significant difference” is an a priori test and it automatically corrects for inflation of type I error rate.
True/ False
FALSE
Right, it is a post hoc test, but it is true that it corrects for inflation of type I error rate.
76.
- Correlation coefficients may vary between -1 and 1.
True/ False
TRUE
77.
- When comparing means, it is essential to do an omnibus F test first, and only if that is significant can you legitimately use either an a priori or a post hoc test.
FALSE
78.
18.A correlation of -1 means that the variables are independent.
True/ False
FALSE
No, they vary in lock step in this case, only in opposite directions.
79.
- An F number reported as “F(4, 88) = 0.8” enables you to conclude with no further testing that the result was nonsignificant.
True/ False
TRUE
Yes: surprisingly perhaps, this is quite true. An F test is a signal to noise ratio test, and if noise and signal are equal (or as here, if noise is greater than signal) you have no hope of significance.
80.
- The usual null hypothesis with correlations is that the population correlation ρ (Greek letter r, “rho”) is nonzero.
True/ False
FALSE
Right, the null hypothesis is that ρ is zero.
81.
- It is important to consider, at an early stage of experimental planning, all of the following: choice of DV and IVs, what your sample(s) should consist of and how you will recruit them, and how the analysis will be conducted.
True/ False
TRUE
82.
- It may be necessary to carry out a power calculation in advance, to ensure a substantial probability of obtaining a significant result if the effect exists.
True/ False
TRUE
Correct, and certainly if you are applying for a grant anywhere.
83.
- If the value of r in your sample is non-zero, the null hypothesis is rejected and the two variables are correlated.
True/ False
FALSE
No: the value of r will (almost) certainly be non-zero due to random variation, even if the null is true.
84.
- The type I error rate is equal to alpha; the type II error rate equals power.
True/ False
FALSE
85.
- When writing a scientific paper or report, you should summarise the outcome(s) of the study in the Introduction section, and include a brief interpretation of these outcomes in the Results section.
True/ False
FALSE
86.
- It is an assumption of most parametric tests involving separate groups that the distributions of the DV within the groups do not depart too far from normality.
True/ False
TRUE
Correct. Note the phrase “within the groups”.
87.
- Suppose a multiple regression includes just one IV, which is a multilevel categorical IV. Then the results of the significance tests carried out in SPSS on the regression coefficients of the dummy variables are equivalent to a set of t tests, each of which compares the values of the DV for one fixed level of the categorical variable, with those at one of the other levels of that variable.
True/ False
TRUE
Yes: this is (Rory thinks) correct, if cumbersome.
88.
- It is an assumption of ANOVA that the variances within the individual groups do not differ by too great a degree from one another.
True/ False
TRUE
Correct, and this is checked by Levene’s test. In practice, there may be ways round this (eg using post hoctests which do not rely on the HOV assumption).
89.
- The type I error rate plus the type II error rate equals 100%.
True/ False
FALSE
No: there is no such simple link between them.
90.
- Multicollinearity in a set of continuous independent variables can be detected by a significance test on the size of the Mahalanobis distance.
True/ False
FALSE
91.
- The basic idea of ANCOVA is one of “signal reduction”: it is used to improve the power of an ordinary ANOVA by using a covariate to reduce the effect size associated with the ANOVA.
True/False
FALSE
No, on two counts: it is a method of noise reduction, and it will certainly do us no good to reduce an effect size.
92.
- In a repeated measure design in which time is the independent variable, it is often advisable to carry out a trend analysis on the values obtained, and in this case checks for sphericity may not be appropriate.
True/ False
TRUE
93.
- The type I error rate is equal to the alpha significance level, and the type II error rate (in percent) plus the power (in percent) equals 100%.
True/ False
TRUE
At last, the correct formulation.
94.
- Multivariate outliers can be detected in a distribution by looking at histograms of each individual variable.
True/ False
FALSE
95.
- Strongly nonnormally distributed data can sometimes be converted into a form suitable for a parametric test by transforming them.
True/ False
TRUE
96.
- It is not possible to enter a categorical (noninterval) variable with fewer than three levels as an IV into a multiple regression unless it is first coded into a number of dichotomous variables.
True/ False
FALSE
97.
- One requirement for the validity of ANCOVA is homogeneity of regression, i.e. if regression lines are plotted separately for each group for the DV on the CV, the slopes of these lines do not significantly differ from one another.
True/ False
TRUE
98.
- In a pretest/posttest design to assess the effect of a treatment, the analysis was carried out using repeated measures ANOVA using the pretest and posttest values as the within subjects variable, and the treatment and control groups as the between subjects variable. The existence of a effect can be determined by seeing whether the interaction between the group variable, and the within subjects variable, is significant.
True/ False
TRUE
Correct. It seems counter-intuitive, but what is essentially a main effect, that of treatment, has to be tested by looking for an interaction. This is because treatment effects involve changes in DV values between pretest and posttest (within subjects effect) compared between groups.
99.
- The following set of coefficients can be input into a oneway ANOVA with five groups to test a difference between two groups of means:
(1, 1, 1, -1, -1).
True/ False
FALSE
100.
- Cronbach’s alpha is a measure of external validity amongst the items of some measuring instrument.
True/ False
FALSE
101.
- In factor analysis, Bartlett’s test of sphericity should be nonsignificant for any factors to exist.
True/ False
FALSE
