GoConqr has a new look! Let us know what you think.

Sign up free

Flashcards

Analysis of Psychometric Data

Hannah Luke Psychology English 136 cards

Cards in this deck (136)

  1. 1Purpose of Reliability AnalysisAssessing internal consistency of a scale
    Scale improvement via exclusion of items
  2. 2Purpose of PCA (Principle Component Analysis)Identifying sub-scales (item/variable groups)
    Variable reduction, re-representation, orthogonalisation
  3. 3Purpose of EFA (Exploratory Factor Analysis)Explaining item responses from underlying 'latent dimensions'
  4. 4Purpose of Cluster AnalysisIdentifying sub-groups of observations (sampling units, subjects); multivariate classification
  5. 5Purpose of RegressionUsing scales for prediction and hypothesis testing
    In particular: Regression for rating-scale data
  6. 6Which size of sample are more likely to have a sampling error?Smaller samples
  7. 7Reliability analysis assumes that items of the scale measure...a single construct
  8. 8Reliability Analysis cannot be interpreted as a 'test of...unidimensionality
  9. 9Classical testing theory states that the assumptions of Reliability Analysis are;The psychological construct exists, and the items of your scale (sub-scale) are designed to measure that particular construct
    Every item measures the true score on the latent variable subject to random error
    Item response = true score + error
  10. 10Adding or averaging responses across items will _______ the accuracy of the true score measurement. Why?Improve
    The errors will cancel each other out
  11. 11Items that measure mostly error should be ____ from the scale in order to _____ reliability.Excluded
    Improve
  12. 12What is the most commonly reported reliability measure?Cronbach's Alpha
  13. 13The variance of the sum scale will be ____ than the sum of the individual item variances if the items measure ____variability across subjects, that is, if they measure _________.greater
    the same
    mostly true score variance
  14. 14We can estimate the proportion of true score variance that is captured by the items by comparing the ....the sum of the item variances with the variance of the sun scale
  15. 15Cronbach's Alpha ranges between ___ and measures _____ High value = ______0 and 1
    internal consistency of a scale
    high consistency
  16. 16For a reasonably useful psychometric measurement, you want to achieve a Cronbach's Alpha of at least...0.7
  17. 17In order for reliability to be assessed, negatively correlated items need to be....reversed
  18. 18raw_alpha = .81 lower = 0.7 upper = 0.91 95% confidence boundary How would this output be interpreted?We can say with 95% confidence for the total population of the sample that the alpha should be between 0.7-0.91.
    Raw Alpha indicates a good level 0.81
  19. 19Define std.alphaAlpha when items are z-standardised
  20. 20Define Guttman's lambda 6Based on squared multiple correlation
  21. 21Define average RAverage inter-item correlation (used to calculation std.alpha)
  22. 22What range of average r would indicate a good scale?Between 0.2-0.4
  23. 23What output would you consider before dropping a scale item?Reliability if an item is dropped: Raw alpha
    Item statistics: raw.r
  24. 24What does the Item Statistics output show?Correlations between individual items and the total scale
  25. 25What does the Frequency Table output show?Distribution of response per item (probabilities of response categories and missing responses)
  26. 26Why would you consider bootstrapping your Reliability Analysis?If you've conducted an analysis that returns parametrically estimated confidence intervals for Cronbachs Alpha it will assume a normal sampling distribution in the population. May not be true so bootstrapping offers an alternative, nonparametric approach.
  27. 27What does bootstrapping do?By repeating a large number of times, you obtain a distribution for the statistic of interest which approximates the population sampling distribution for that statistic, which can be used to derive confidence intervals.
  28. 28Sampling with replacement (in bootstrapping) means...?any of the original subjects can appear several times, or indeed not at all
  29. 29Doing Reliability Analysis on items scored on different scales will impact which outputs? (e.g. some items are scored 1-5 and some are scored 1-10)Raw Alpha and Scale Total (final scale composite score)
  30. 30If items are scored on different scales you need to ___ them before conducting analysis.standardise
  31. 31What would violate the Tau-Equivalence assumption?Scales scored differently. These different scales imply different 'weights' per item.
  32. 32Other than differently scored items when might you standardise scores before analysis?If subjects use vastly different ranges of the same rating scales across the different items
  33. 33What is the most commonly used method to standardise scores?Z-score standardisation
  34. 34High reliability in the measurement of the construct of interest does not guarantee..?validity
  35. 35Reliable scales will ____ statistical powerimprove
  36. 36Items that correlated strongly with one another measure more or less the same construct, and thus form a _____ within our measurement instrumentgroup or sub-scale
  37. 37In essence, Principal Component Analysis performs data compression. Define thisReduce a large number of items into a smaller number of 'latent' dimensions' such that items that are strongly correlated load on the same latent dimension
  38. 38Describe the first analytical step of PCAStart with a set of items measured on a number of subjects (as with reliability analysis)
  39. 39Describe the second analytical step of PCADetermine the number of latent dimensions (components/factors) you want to "extract" from the data
  40. 40Describe some criteria that may help you decide how many latent dimensions to extract in PCA?Start by extracting as many dimensions as items
    Extract dimensions with Eigenvalues > 1
    Extract sufficient dimensions to retain 80% of the data
  41. 41Describe the third analytical step of PCARotate and interpret (label) the extracted latent dimensions
  42. 42Describe the fourth analytical step of PCADetermine factor scores
  43. 43Define factor scoresStandardised liner combinations (weighted sums) of the original items representing each latent dimension
  44. 44Why are factor scores useful in PCA?If you want to use those dimensions as IVs or DVS in later analyses
    For example, compressed information can be used as predictor variables in linear regression
  45. 45PCA extracts ______ latent dimensionsorthogonal (i.e. uncorrelated)
  46. 46What are the advantages of orthogonal extraction?Orthogonal dimensions are useful if you want to use them are predictors in regression analysis.
    If you used correlated predictors instead it will often yield unreliable regression parameters because their contributions to explaining a DV are difficult to disentangle
  47. 47What are the disadvantages of orthogonal extraction?Orthogonal dimensions may not always make a lot of sense and are potentially difficult to intrepret
  48. 48Each item can be represented as a _____linear combination of the latent dimensions
  49. 49If there are many latent dimensions as original items and if every item is maximally loaded on its own dimension and minimally loaded on other dimensions. Then the error would be?Zero
    The overall variance is fully accounted for
  50. 50Eigenvalues express...?the amount of variance explained in the number-of-items units
  51. 51Eigenvalues per factor are dependent on...?rotation
    A rotated vs unrotated solution would reveal different distribution of Eigenvalues across the factors identified
  52. 52What is the purpose of Parallel Analysis?To determine the number of PCs or factors to extract in PCA.
  53. 53What is the first step of Parallel Analysis?From the given data, simulate or permutate how the distribution of Eigenvalues per successive PC/factor would look like if the data was completely random
  54. 54What is the second step of Parallel Analysis?Compare the actual distribution of Eigenvalues with the randomly generated distribution of Eigenvalues (using a scree plot)
  55. 55What is the third step of Parallel Analysis?Set the number of PCs/factors to extract equal to the number of PCs/factors that have Eigenvalues greater than their (simulated or permutated) random counterparts
  56. 56Define Rotation of factorsA transformation of the factor vectors (comparable with turning a coordinate system in space) in order to reduce ambiguity
  57. 57Define varimax rotationMaximise the variance of the squared factor loadings (column) on all items (rows) in a matrix of loading
  58. 58A varimax solution yields results that make it as easy as possible to identify....?each item with a single factor
  59. 59Define quartimax rotationMinimises the number of factors needed to represent each item. This type of rotation often generates a general factor on which most items are loaded to a high or medium degree
  60. 60Define equamax rotationA compromise between varimax and quartimax rotation. Not necessarily the best for interpretation
  61. 61Define CommunalitiesIndicate for each individual item on the original scale how much variance is captured by the extracted factors
  62. 62Whereas in PCA, items/variables are assumed to linearly combine into factors, in EFA factors are assumed to explain...individual item responses
  63. 63To determine the number of PCs/factors to extract in EFA we should look at what?Root Mean Squared Error of Approximation (RMSEA)
  64. 64What is an alternative method of determining factor extraction for EFA than RMSEA?Baussian Information Criterion
  65. 65What RMSEA score are we considering for factor extraction?The lower RMSEA the better
    RMSEA <.06 usually considered OK
  66. 66If RMSEA = 0, what could that indicate?Overfitting, too many factors in the model so the model starts to ascribe meaning to random noise
  67. 67What is the first step of EFA?Determine the number of factors to extract
  68. 68What is the second step of EFA?Determine the Principal Components
  69. 69What is the third step of EFA?Calculate the loadings of factors to Principal Components
  70. 70What the fourth step of EFA?Interpretation of components
  71. 71What is the typical lower end cut off for loading scores?0.3 but depends on the size of sample. Smaller samples may need to focus on stronger loadings i.e. higher cut off
  72. 72What is the fifth step of EFA?Calculating Factor Scores
    Add (properly labelled) standardised factor scores to the data if we want to use the extracted principal components in later analyses
  73. 73What is an alternative method to determining the number of factors to extract inBaussain Information Criterion
  74. 74What score of RMSEA do you consider for factor extraction?The lower RMSEA the better, RMSEA < .06 usually considered OK
  75. 75What does an RMSEA = 0 potentially indicate?Overfitting, too many factors in the model so the model starts to ascribe meaning to random noise
  76. 76What is the first step in EFA?Determine the number of factors to extract
  77. 77What is the second step in EFA?Determine the principal components
  78. 78What is the third step in EFA?Calculate the loadings of factors in PCs
  79. 79What is the typical cut off for loading scores?0.3 but depends on the size of sample
  80. 80Why would the cut off for loading scores change depending on sample size?Smaller samples may need to focus on stronger loading which means higher cut offs
  81. 81What is the fourth step in EFA?Interpretation of components
  82. 82What is the fifth step of EFA?Add (properly labelled) standardised factor scores to data if we want to use the extracted principal components in later analyses
  83. 83Each weight (in factor scores) refers to...the loading of a given item on a given factor in proportion to the sum of all item-loadings per factor
  84. 84Compared to PCA, EFA has a stronger focus on...understanding and explanation of variable relationships
  85. 85In PCA, the components are calculated as...linear combinations of the original variables/items (i.e. items contribute to components)
  86. 86In EFA, the original variables/items are defined as...linear combinations of the factors (i.e. factors explain individual item response)
  87. 87PCA extracts _____ components; therefore it should only be used in combination with ___ rotation methods such as...orthogonal
    orthogonal
    varimax, quartimax and equamax
  88. 88In EFA (unlike PCA), you can use ___ rotation methods such as _____, meaning that factors can be ____.oblique
    oblimin or promax
    correlatated
  89. 89In EFA, the loadings should be interpreted as...extracted-factor contributions to the explanation of individual item variance
  90. 90What kind of technique is k-means clustering?unsupervised machine based learning
  91. 91What does k mean clustering find?Clusters of observations in a multivariate space such that within-cluster similarities are maximised and between-cluster similarities are minimised
  92. 92To quantify space in clustering you consider (conceptually) that similarity is the inverse of ...distance
  93. 93The most common unit of distance used in clustering is...Euclidean Distance
  94. 94Define Euclidean DistanceThe ordinary straight-line distance between two points in a co-ordinate system
  95. 95Euclidean Distance is calculated usingPythagorean Theorem
  96. 96How would you consider the standard deviation of a variable in terms of Euclidean Distance?The standard deviation of a variable is the average Euclidean Distance between the mean of that variable and the individual data points
  97. 97What are the two cautions to consider for Euclidean Distance?Euclidean Distance is scale-sensitive (dependent on measurement unit)
    Clustering based on Euclidean Distance is sensitive to outliers
  98. 98What's the first step of K-means clustering?Randomly choose K cases as initial centroids
  99. 99What's the second step of K-means clusteringAssign cases to centroids such that the sum of distances is minimised (nearest cases to the centroids)
  100. 100What is the third step of K-means clustering?Determine resulting new centroids (previously not in the sample)
  101. 101What is the fourth step of K-means clustering?Repeat steps 2 and 3 until clusters don't change anymore or max number of iterations is reached
  102. 102How would K-mean clustering starting conditions affect results?Different (random) choices of initial centroids often yield different cluster solutions (i.e. different groups of subjects)
  103. 103List the two clustering quality criteria optionsch - Calinski-Harabasz Index
    asw - Average Silhouette Width
  104. 104Calinski-Harabasz index looks for...larger/more evenly-sized clusters
  105. 105Average Silhouette Width measures...how tightly grouped all the points per cluster are
    Comes up with smaller/less evenly-sized clusters
  106. 106PCA factor scores are standardised. What is the advantage of this with K-mean clustering?Good for clustering if we assume that each component is "equally important"
  107. 107PCA factor scores are standardised. What is the disadvantage of this with K-mean clustering?If we want components that account for more variance in the original items to have a higher weight in the classification of cases
  108. 108Define Jaccard similarityMeasures how well each cluster from the best clustering solution replicates over the B random resamples from the original subjects
  109. 109Average Jaccard similarity values greater than ___ are considered to indicate reasonably robust clusters that are unlikely to 'dissolve'0.6
  110. 110K-means clustering is sensitive to...Scaling (determines the weight of each clustering dimension)
    Potential outliers (check these first)
    Starting conditions (initial cluster centroids)
  111. 111Which type of regression would you conduct on binary data?Logistic regression
  112. 112Which kind of regression would you conduct on rank data?Ordinal logistic regression
  113. 113What is the goal for linear regression?Predict a continuous DV from a continuous IV assuming a linear relationship between the two
  114. 114The 'best fitting' through an x and y data cloud is one that minimises...the residuals (prediction errors)
  115. 115Assumptions for linear regressionVariables are measured on at least interval scale
    Linearity/additivity
    Homoescedasticity
    Normality of residuals
  116. 116What are some of the various applications for multiple linear regression?Prediction
    Confirmatory testing of individual parameters
    Exploratory 'model selection'
  117. 117The general purpose of multiple regression is to...learn more about the linear relationship between several independent variables and a single dependent variable
  118. 118Multiple regression assumes that...the predictor variables are independent of one another (i.e. uncorrelated)
  119. 119When predictors are on different scales, these will need to be ____ before regression can runcentered
  120. 120Centring the predictors around their mean yields...an intercept that has a meaningful interpretation
  121. 121Adjusted R-squared is equal to...R-squared plus penalty for increasing model parameters
  122. 122Define Generalised Linear ModelsAn extension of linear models that allow for specification of distribution and link functions to accommodate a variety of different data types
  123. 123Define varianceDistribution of residuals
  124. 124Define Linkapplies a transformation to the model parameters and determines the interpretation of model coefficients
  125. 125Probability ranges from...0 to 1
  126. 126Odds ranges from0 to infinity
  127. 127Logits range between...-infinity to +infinity
  128. 128The logit transform fails if probability equals...0 or 1
  129. 129In the "middle" probability range, small changes in P imply...small changes in logit
  130. 130When probabilities approach logical boundaries, small changes in P imply...large changes in logit
  131. 131Applying a logistic function to P(y) is pretty much the same as applying...a linear function to the log odds of P(y) which is what binary logistic regression does
  132. 132Instead of adjusted R-squared in standard linear regression, glm reports...Akaike Information Criterion (AIC)
  133. 133What is Akaike Information Criterion (AIC)?A goodness of fit measure
  134. 134Lower values of AIC mean...better fit
  135. 135True or False? AIC cannot be interpreted in an absolute senseTrue
  136. 136What do Threshold coefficients indicate?The (estimated) average distribution of scale points in the sample

Information

Description
Psychology (Fourth Year Honours) Flashcards on Analysis of Psychometric Data, created by Hannah Luke on 25/11/2022.
Created by
Hannah Luke
Views
0
Rating

Feedback

What is it about?

What do you think of the new look?

How do you like it?