Questão 1
Questão
What is the purpose of performing a linear regression analysis?
Responda
-
To identify potential outliers in the data
-
To fit the data to a model that defines y as a function or 2 or more variables
-
To determine the dependence of a dependent variable on a predictor/independent variable
-
To perform multiple comparisons whilst controlling overall type 1 error rate
-
To derive robust confidence intervals
Questão 2
Questão
Which axis does the dependent variable go on?
Questão 3
Questão
What does the mean of the x and y values give you in a linear regression analysis?
Responda
-
The size of the force which the points exert on the line of best fit
-
The leverage of those data points
-
The fit and slope of the model
-
The centre of gravity and pivot point of the data
Questão 4
Questão
What does the R-squared value represent?
Responda
-
How well the model fits the data (0 - 1)
-
The slope coefficient
-
The distribution of the residuals
-
The level of multicolinearity in the model
Questão 5
Questão
What does an R-squared value of 0.068 and a slope coefficient (b1) value of 0.12 mean?
Responda
-
The model can explain 68% of the data and for every unit of independent variable, the dependent variable goes up 12 units
-
The fit of the model to the data is 0.12% and the influence that the data points have on the model is 0.68%
-
The data points have an influence of 68% on the model and 12% on the outcome
-
The model can explain 6.8% of the data and for every unit of independent variable, the dependent variable goes up 0.12 units
Questão 6
Questão
In order to identify potential outliers:
Responda
-
Standardised residual >2 is worth checking, if more than 5% of the residuals >2 may indicate model is a poor fit
-
Standardised residual >3 is worth checking, if more than 5% of the residuals >2 may indicate that the model is a poor fit
-
Standardised residual >2.5 is worth checking, if more than 5% of the residuals >2 may indicate that the model is a poor fit
-
Standardised residual >3 is worth checking, if more than 10% of the residuals >2 may indicate that the model is a poor fit
Questão 7
Questão
What does Cook's distance tell us when performing model diagnostics to see if the regression model is stable or biased by a few cases?
Responda
-
influence of data point on predicted values (0 = no influence, 1 = complete influence)
-
standardised measures of how much each element of the model would change if data point was removed (values >1 = substantial influence)
-
how susceptible the mean is to being biased by the outliers present in the data
-
measure of overall influence of each individual data point on the overall model (>1 = concern)
Questão 8
Questão
What does the Leverage value tell us when performing model diagnostics to see if the regression model is stable or biased by a few cases?
Responda
-
influence of data point on predicted values (0 = no influence, 1 = complete influence)
-
standardised measures of how much each element of the model would change if data point was removed (values > 1 = substantial influence)
-
measure of overall influence of each individual data point on the overall model (> 1 = concern)
-
precisely how large the standardised residuals are
Questão 9
Questão
With regard to model diagnostics, what do the DFFit and DFBeta values tell us about the data model?
Responda
-
they are standardised measures of how much each element of the model would change if that data point was removed (values > 1 = substantial influence)
-
they indicate the influence of that data point on predicted values (0 = no influence, 1 = complete influence)
-
whether or not the standardised residuals are worth checking and if they indicate that the model is a poor fit
-
they summarise the equation: 2(k+1)/n where k = number of predictors and n = number of data points
Questão 10
Questão
With regard to the model diagnostic called the Leverage value, what defines whether or not the data point is worth investigating?
Responda
-
if >2(k+1)/n where k = number of predictors (2 for simple linear regression) and n = number of data points
-
if >2(k+1)/n where k = number of predictors (1 for simple linear regression) and n = number of data points
-
if >2(K+1)/n where k = number of data points and n = number of predictors (1 for simple linear regression)
-
if >n(k+1)/2 where k = number of predictors (1 for simple linear regression) and n = number of data points
-
if >2(n+1)/k where k = number of predictors (1 for simple linear regression) and n = number of data points
Questão 11
Questão
Multiple linear regression does what?
Responda
-
fits the data to a model that defines y as a function of 2 or more variables - determines the effect of an independent variable on the dependent variable taking account of other variables
-
provides an analysis of variance and determines if an interaction is present in the data
-
determines the dependence of a dependent variable on a predictor/independent variable and allows outliers to be identified from x, y plot or from standardised residual plot
Questão 12
Questão
With regard to multiple linear regression, what is the correct form of the equation for the model which is fitted? (all of the numbers are technically subscript)
Questão 13
Questão
What does the F-ratio represent?
Responda
-
the average variability due to the model divided by the average variability due to the residuals
-
the unexplained variability divided by the variability due to the model
-
the signal to noise ratio multiplied by the number of data points
-
the variance in the model divided by the R-squared value
Questão 14
Questão
With regard to multiple linear regression, whenever you fit a predictor variable, that takes up...
Responda
-
one slope parameter
-
two degrees of freedom
-
one degree of freedom
-
one R-squared value
Questão 15
Questão
As colinearity increases what effect does this have?
Responda
-
standard errors of b coefficients decrease therefore confidence increases
-
limits F-ratio value and variance inflation factor
-
coefficients become stable
-
standard errors of b coefficients increase and therefore confidence decreases
Questão 16
Questão
How do you interpret the variance inflation factor (VIF) when assessing multicolinearity?
Responda
-
A VIF > 5 or an avereage VIF > 2 is problematic
-
A VIF > 10 or an average VIF > 1 is problematic
-
A VIF > 2 or an average VIF > 1 is problematic
-
A VIF > 10 or an average VIF > 2 is problematic
Questão 17
Questão
How do you interpret the tolerance factor when assessing multicolinearity?
Responda
-
< 5 is problematic
-
< 10 is problematic
-
< 2 is problematic
-
< 0.1 is problematic
-
< 1 is problematic
Questão 18
Questão
When does multicolinearity truly pose a problem?
Responda
-
when predicting y using the multiple regression equation
-
when you want to look inside the model at the effect of individual predictors
-
when you want to perform separate correlations for each x variable
-
when you want to quantify the relationship between an independent and dependent variable
Questão 19
Questão
How do you help solve the problem of multicolinearity?
Responda
-
always take a colinear variable out
-
combine predictors into a single predictor (as long as it makes biological sense)
-
rely on automatic variable selection
-
remove all outliers
Questão 20
Questão
With regard to hierarchical multiple regression, what value do you use when comparing new model to previous model?
Questão 21
Questão
For multiple linear regression assumptions, what must the variables be?
Responda
-
dependent variables = quantitative or categorical
predictor variable = qualitative and continuous
-
dependent variables = qualitative
predictor variable = continuous
-
dependent variables = qualitative and continuous
predictor variable = quantitative or categorical
-
dependent variables = continuous or categorical
predictor variable = quantitative or categorical
Questão 22
Questão
When considering multiple linear regression assumptions, how do you assess the independence of the residuals?
Questão 23
Questão
For multiple linear regression, how large should the sample size be?
Questão 24
Questão
What would an interaction among predictors look like in the form of an equation?
Responda
-
effect of height + effect of weight = overall effect on SBP
-
effect of height + overall effect on SBP = effect of weight
-
effect of height x effect of weight = overall effect on SBP
Questão 25
Questão
What is simple linear regression equal to?
Responda
-
paired t-test
-
unpaired t-test
-
unpaired, two-tailed t-test
-
paired, one-tailed t-test
Questão 26
Questão
A one-way anova is the same as what?
Responda
-
unpaired t-test
-
paired t-test
-
simple linear regression
-
multiple regression
Questão 27
Questão
What does a one-way ANOVA do?
Responda
-
analyses how much of the overall variance can be explained by variation between group means compared to the unexplained variation within a group
-
fits data to a model that defines y as a function of 2 or more variables
-
performs separate correlations for each x variable
-
determines the dependence of a dependent variable on a predictor/independent variable
Questão 28
Questão
What does the total variability equal?
Responda
-
total squares divided by the degrees of freedom
-
the F-ratio
-
the difference between each individual data point and the overall mean
-
error mean squares divided by degrees of freedom
Questão 29
Responda
-
higher the larger the difference of the group means from the overall mean and smaller the amount of random variability
-
lower the larger the difference of the group means from the overall mean and smaller the amount of random variability
-
higher the larger the difference of the group means from the overall mean and larger the amount of random variability
-
higher the smaller the difference of the group means from the overall mean and smaller the amount of random variability
Questão 30
Questão
When is the ANOVA most robust to deviations from normality and equality of variance?
Responda
-
when effect size is large
-
when the F-ratio is high
-
when the degrees of freedom are greater than 10
-
when the group sizes are equal
Questão 31
Questão
If group sizes are unequal and equality of variance is not met then which correction do you use?
Responda
-
Games-Howell's
-
Durbin-Watson's
-
Gabriel's
-
Tukey's
-
Welch's
Questão 32
Questão
What are post-hoc tests used for?
Responda
-
performing multiple comparisons whilst controlling overall type 2 error rate
-
performing multiple comparisons whilst controlling overall type 1 error rate
-
when there is a specific hypothesis to be tested
-
when group size is not equal
Questão 33
Questão
You use Tukey's test when which of the following is true? (multiple correct answers)
Responda
-
sample sizes are unequal
-
sample sizes are equal
-
you require good trade-off between type 1 and type 2 errors
-
you are interested in comparing all groups vs a single control group
-
you wish to cut down on the number of comparisons that you make
Questão 34
Questão
When would you use Bonferroni as a post-hoc test? (multiple correct answers)
Responda
-
when you don't need a high level of confidence
-
when you aren't performing multiple comparisons
-
when you require a conservative test
-
when you need a high level of confidence
-
when sample sizes are equal
Questão 35
Questão
When would you use Dunnet's as a post-hoc test? (multiple correct answers)
Responda
-
when interested in comparing all groups versus a single control group
-
when sample sizes are equal
-
when you want to cut down comparisons
-
when you want a good trade-off between type 1 and type 2 errors
Questão 36
Questão
Separate, unpaired t-tests to do comparisons will increase your chance of getting what?
Responda
-
a false -ve
-
a type 2 error
-
a false +ve
-
biased data
Questão 37
Questão
If you have a sample which has an n number of 10 and a sample with an n number of 12, which post hoc test should you use?
Responda
-
Gabriel's
-
Hochberg's GT2
-
Games-Howell
-
Tukey
Questão 38
Questão
If there is any doubt about equality of variance then which post-hoc test should you use?
Responda
-
Gabriel's
-
Hochberg's GT2
-
Games-Howell
-
Sidak
Questão 39
Questão
Complete this statement relating to planned contrasts:
Always _________________________ than number of groups
Questão 40
Questão
When doing orthogonal contrasts, the contrasts are independent so you can... (multiple correct answers)
Responda
-
enter weights for most of the variables
-
trust p-values as you aren't inflating the type 1 error rate
-
ignore the F-ratio value and R-squared value
-
not worry about doing any corrections for multiple comparisons
Questão 41
Questão
"tests for trends in the data, which cannot be obtained directly using post-hoc tests, when there is a logical order to the groups entered"
To what is this statement referring to?
Responda
-
Planned contrasts
-
Orthogonal contrasts
-
Polynomial contrasts
Questão 42
Questão
An independent factorial ANOVA does what?
Responda
-
each level of one factor is tested against at least one level of the other
-
performs multiple comparisons whilst controlling overall type 1 error rate
-
divides total variability in the data set into different sources
-
each level of one factor is tested at each level of the other
Questão 43
Questão
Sidak is the best correction for what?
Questão 44
Questão
Standard contrasts and post hoc tests are only available to examine main effects and are therefore most useful when:
Questão 45
Questão
A p value of less than 0.5 means that....
Responda
-
there is a less than 0.5% chance of committing a type 1error
-
there is a less than 5% chance of committing a type 2 error
-
there is a less than 5% chance of committing a type 1 error
-
there is a less than 0.5% chance of committing a type 2 error
Questão 46
Questão
Standard error the proportion equals....
Questão 47
Questão 48
Questão
Power can be increased by....
Responda
-
increasing effect size. decreasing random variation. decreasing sample size.
-
increasing effect size. increasing random variation. increasing sample size
-
decreasing effect size. increasing random variation. increasing sample size
-
increasing effect size. decreasing random variation. increasing sample size
Questão 49
Questão
This gives you a standardised effect size for a difference between means, what is it called?
Questão 50
Questão
how do you calculate expected frequency?
Responda
-
(row total + column total)/overall total
-
(row total - column total)/overall total
-
(row total x column total)/overall total
Questão 51
Questão
How do you calculate degrees of freedom from a contingency table?
Responda
-
df = (rows - 1) x (columns -1)
-
df = (rows + 1) x (columns +1)
-
df = (rows - 1) / (columns -1)
-
df = (rows x 2) + (columns x 2)
Questão 52
Questão
With regard to categorical data - what must be satisfied in order for the analysis to be reliable?
Responda
-
The assumption that at least 50% of expected frequency must be more than or equal to 5
-
Dunnet's test
-
The assumption that at least 80% of expected frequency must be more than or equal to 5
-
The same assumptions as multiple linear regression
Questão 53
Questão
Which graph indicates an interaction?
Questão 54
Questão
What does simple effects analysis do? (multiple correct answers)
Responda
-
probes where a certain effect is happening
-
performs an ANOVA to allow you to reject/accept a null hypothesis
-
analyses the differences between levels of one variable
-
performs multiple comparisons whilst controlling overall type 2 error rate
Questão 55
Questão
Which multiple comparison correction should you choose after simple effects analysis in order to control the overall type 1 error rate?
Questão 56
Questão
Repeated measures ANOVA requires the data to have/be:
Questão 57
Questão
What is the definition of sphericity?
Responda
-
“noise” in the relationship between the independent variables and the dependent variable is the same across all values of the independent variables
-
equality of differences between linked values in each group
-
well-modeled by a normal distribution and likely for a random variable underlying the data set to be normally distributed
-
residuals are (roughly) normal and (approximately) independently distributed with a mean of 0 and some constant variance
Questão 58
Questão
Which test provides a fix for sphericity?
Responda
-
Welch's correction
-
Games-Howell test
-
Gabriel's test
-
Mauchy's test
Questão 59
Questão
How can we adjust the degrees of freedom and change the significance level associated with the F-statistic?
Questão 60
Questão
Which post hoc test is most robust and most conservative for a repeated measures ANOVA?
Responda
-
Sidak
-
Tukey
-
Dunnets
-
Bonferroni
Questão 61
Questão
If parametric assumptions are in doubt, we must use the non-parametric equivalent of a single factor repeated measures ANOVA which is:
Responda
-
Durbin-Watson test
-
Friedman test
-
Gabriel's test
-
Hochberg's GT2 test
Questão 62
Questão
For which analysis do BOTH the equality of variance assumption and sphericity assumption apply?
Responda
-
Non-linear regression
-
Two-way ANOVA
-
Independent ANOVA
-
Mixed model ANOVA
-
Repeated measures ANOVA
Questão 63
Questão
In terms of polynomial regression, what happens if you add further terms to the polynomial? (multiple correct answers)
Responda
-
the fit will automatically improve
-
there is a risk of over-fitting the model
-
the significance level associated with the F-statistic changes
-
the R-squared value will increase
Questão 64
Questão
In terms of nonlinear regression, why would you want to try multiple starting parameters?
Responda
-
to ensure that the interaction between the variables is taken into account
-
to ensure that the computer has found the global minimum
-
to ensure that the computer has found the local minimum
-
to ensure that an accurate scientific relationship is found
Questão 65
Questão
How would you calculate the Sum of Squares (SS)?
Responda
-
add all the standard deviations together and square that value
-
square the mean from each sample and add those together
-
square each standard deviation and add them all together
-
square each standard deviation and add this to the variance
Questão 66
Questão
Variance is calculated by doing what?
Responda
-
dividing the standard deviations by the degrees of freedom
-
dividing the sum of squares by the degrees of freedom
-
multiplying the degrees of freedom by the mean
-
multiplying the standard deviations by the sum of squares
Questão 67
Questão
How do we define the normal distribution curve?
Responda
-
the population mean is the height and the sum of squares is the distance from the midline of the curve to the edge
-
the variance is the height and the population mean is the distance from the midline of the curve to the edge
-
the population standard deviation is the height and the population mean is the distance from the midline of the curve to the edge
-
the population mean is the height and the population standard deviation is the distance from the midline of the curve to the edge
Questão 68
Questão
How do you calculate a z-score?
Responda
-
(x - mean) /sd
-
(x - sd)/mean
-
(mean-x)/sd
-
(x + mean)/sd
Questão 69
Questão
Choose all of the correct statements
Responda
-
approximately 99% of normally-distributed values lie between +- 2 sds from the mean
-
approximately 95% of normally-distributed values lie between +-2 sds from the mean
-
approximately 99.9% of normally-distributed values lie between +- 2.6 sds from the mean
-
approximately 99% of normally-distributed values lie between +- 2.6 sds from the mean
-
approximately 99.9% of normally-distributed values lie between +- 3 sds from the mean
-
approximately 95% of normally-distributed values lie between +- 3 sds from the mean
Questão 70
Questão
How do you calculate SEM and therefore, confidence intervals?
Responda
-
SEM = sd x square root of n and therefore a 95% CI would be +- 1.96 x SEM
-
SEM = sd/square root of n and therefore a 95% CI would be +- 3 x SEM
-
SEM = sd x square root of n and therefore a 95% CI would be +- 2.6 x SEM
-
SEM = sd/square root of n and therefore a 95% CI would be +- 1.96 x SEM
Questão 71
Questão
Which statement is true?
Responda
-
P < 0.05 means that 5% of the results arose by chance if the null hypothesis is true
-
P < 0.05 means <5% probability of the results arising by chance if the null hypothesis is true
-
P < 0.05 means <0.05% probability of the results arising by chance if the null hypothesis is true
-
P < 0.05 means that <0.5% probability of the results arising by chance if the null hypothesis is true
Questão 72
Questão
Choose the correct statements
Responda
-
type 1 error rate is conventionally set to 5% ( P < 0.05)
-
type 2 error rate is conventionally set to 5% ( P < 0.05)
-
type 1 error rate = 1 - power
-
type 2 error rate = 1 - power
-
if you accept a statistical power of 80% it will mean that you have a type 2 error rate of 20%
-
if you accept a statistical power of 80% it will mean that you have a type 1 error rate of 20%
Questão 73
Questão
What happens if you design an experiment with 3 groups and are tempted to test for differences between the means using 3 separate t-tests? (multiple correct answers)
Responda
-
you will increase the chance of making a type 2 error
-
you will increase the chance of making a type 1 error
-
you will inflate your p-value
-
you will decrease your p-value
Questão 74
Questão
Which statements are correct regarding the Pearson Correlation Coefficient?
Responda
-
+- 0.5 is a large effect
-
+- 0.1 is a small effect
-
+- 1 is a small effect
-
it measures how close the data points are to a straight line that best describes the linear relationship
-
r = +0.1 refers to a perfect straight line with a positive slope
-
r = -1 refers to a perfect straight line with a negative slope
Questão 75
Questão
How is the line of best fit created in simple linear regression?
Responda
-
by minimising the total sum of squares
-
by minimising the sum of squares of the residuals
-
by creating an equation which fits the model best
-
by entering the data into the computer in Hierarchical form
Questão 76
Questão
How do you calculate R squared?
Responda
-
1 - (SS of the residuals/total SS)
-
1 + (SS of the residuals/total SS)
-
1 - (total SS/SS of the residuals)
-
1 + (total SS/SS of the residuals)
Questão 77
Questão
Shapiro Wilk's test is used to...
Responda
-
check for sphericity
-
correct degrees of freedom
-
ascertain that residuals are random and normally distributed
-
minimise the sum of squares of the residuals
Questão 78
Questão
When entering more than 2 categories as dummy variables... (multiple correct answers)
Responda
-
the thing that you're comparing the baseline to gets a 1
-
the thing that you're comparing the baseline to gets a 0.1
-
1 fewer dummy variables than number of categories
-
baseline condition gets a value of 0
-
baseline condition gets a value of 1.5
Questão 79
Questão
Bonferroni test on its own - the p-values need to be less than what to claim significance?
Questão 80
Questão
Normally distributed variables X and Y are significantly correlated with a p level of 0.006 and a Pearson’s correlation coefficient of 0.468. Approximately how much of the variability in X and Y can be explained by this correlation?