Flashcards
Stats Flashcards SEM1
Cards in this deck (48)
- 1Bootstrapping: used for... (2 things)(1) Small skewed data sets
(2) statistics without a known distribution - 2Bootstrapping: purpose of testCan determine confidence intervals - assess whether dataset contains zero if it doesn't we can be confident there is a difference between means
- 3Bootstrapping: process (5 steps)(1) randomly drawn subsamples drawn from larger sample repeatedly (sample with replacement) (approx. 2000 times); (2) scores plotted on frequency distribution, (3) determine if it looks normal/biased, (4) calculate CI (remove 2.5th and 97.5th percentiles), (5) if doesn't contain zero, can determine difference between means
- 4Bootstrapping: interpretationIf CI contains zero - no difference between 2 groups. if doesn't contain zero - difference between groups
- 5Bootstrapping: problems- Can't generalise to real world (only study)
- Requires reasonable sample size >5
- Can't correct for poorly designed studies and biased sampling - 6Chi Sq. 1 way goodness of fit test - purposeTests how well observed distribution of scores matches a distribution expected by CHANCE
- 7Chi Sq. 1 way goodness of fit test - processchance = total counts / number of categories
- 8Chi sq. 1 way goodness of fit test - interpretationobserved distribution of scores differs significantly from what would be expected by chance
- 9Chi Sq. 2 way contingency table - purposedetermines whether there is an ASSOCIATION between 2 categorical/ordinal variables
- 10Chi sq. 2 way contingency table - process (if MEF assumption violated what test is used? If not 2x2 how is the effect size given? If 2x2 how is the effect size given?)if MEF assumption is violated - use Kolmogorov Smirnov test
not 2x2 = Kramer's v
2x2 = Fishers exact - 11Chi sq. 2 way contingency table - interpretationobserved chi sq. is compared to critical value (df) independent = no effect, interdependent = an effect
- 12Fischer Exact test: purpose and interpretationPurpose: used for 2x2 designs when MEF assumption underlying chi sq. can't be met.
Interpretation: same interpretation as chi sq. reported as Fischer Exact p<.05 - 13Kappa test: purposePurpose: examines inter-rater reliability in situations involving 2 interdependent raters (estimates degree of correspondence between independent rates)
- 14Kappa test: process and interpretationProcess: K = [agreement-chance]/1 -chance
Interpretation: K = kappa statistic (want it to be higher)
7 = great, 5 = good, 3 = fair - 15Binomial test: purposeUsed to test for the probability of binary outcomes (e.g. L vs R turn) - likelihood of getting numbers in random samples
- 16Binomial test: interpretation and processProcess: Z approximation to binomial
Interpretation: interpretation of Z score, if it is significant means that it is probable to occur in reality - 17Proportion Difference Tests: purposeParametric test. Examines whether 2 samples are drawn from population with the same proportion of a given characteristic (e.g. are males more likely to smoke than females?)
- 18Proportion Difference Tests: process and interpretationProcess: These are easier to interpret than chi-sq as they provide a z score and test statistic as well as being able to calculate CI
Interpretation: interpretation of Z score and test statistic (difference between proportion is clear - see significant, if sig. there is a difference) - 19Poisson Tests: purposetests the probability of event flow in equal sampling units - seeing where things occur most in an area (intervals of time/space) i.e. number of customers coming in at time of day
- 20Poisson Tests: processInterested in only 1 event instead of binary events
observe the mean rate per unit time or area and the actual rate observed - 21Poisson Test: interpretationhigher the p value more likely the even will occur in that time/space, 0.03 not that likely
- 22McNemar test: purposerepeated test that examines whether the probability of a given event is more or less likely in the sample sample over time - 2 TIME POINTS
- 23Cochran's Q: purposeRepeated test that examines the probability of an event less likely in the same sample of people when assessed at a different point in time or in relation to a different choice - 3 TIME POINTS (or more!)
- 24McNemar & Cochran's Q - processCan use together: Cochran's Q for 3 time points to see if any significance, McNemear for 2 time points to work out where significance lies
- 25Problems for McNemar & Cochran's Qcan't do a time x outcome chi.sq test because the data is measured at different time points and the data are non-indenpdenent (collected from the same people)
- 26Logistic regression - purposedependent measure is binary
aim is to determine the best combination of variables which correctly classify cases according to their membership in the two categories defined by dependent measures (0,1) or (1,2) - 27Logistic Regression - processIdentify significant predictors (Walk statistic - t-test, p value); classification table identifies % of correctly classified cases; check wild (higher) and sig. (<.05); Exp(B) - odds ratio - indicates the change in p(E) resulting from a 1 unit change in value of that predictor variable score - how much more likely you're to do something with unit increase
- 28LFemale coded 2, male coded 1. going from 1 to 2 has 18.94 increase so feels are 18.94 times greater to experience neglect
- 29Simple Linear Regression - purpose and interpretationpurpose - determines lines of best fit, minimises residuals by getting close to actual value
interpretation - want R squared (F) to be larger because that indicates the model explains more of the variance
70% = good, 30% ok - 30Multiple Regression - typesStepwise: statistical logic (i.e. highest first)
Hierarchical: theoretical logic
Standard: all variables at once - 31Mediation - purposeeffect of A carried to C via B
A --> B --> C (C dependent, A and B predictors) - 32Mediation - requirements & warningall variables must be significantly correlated, of similar magnitude and in same direction
beware of suppression effect - A-->C is bad (negative) - 33MBaron & Kenny Method. 2 regressions (A predicting C, A+B predicting C)
determines predictive power of A the B is present
same = no change
partial (4)
fully mediated = 0 or negative - 34Moderation - purposeAxB - the simple effect of A on C varies according to B (relationship between A and C differs according to level of B)
- 35Moderation - processMeans of A and B. Multiply to get product term. Hierarchical regression (interaction A*B). Not sig = no moderation, sig = plot effects of predictions to see interaction
- 36Moderation - interpretation & assumptionsnot significant = no moderation
significant = see interactions
assumptions: residuals are equally distributed, check heteroscedascity, normality, non-linearly, non-serial error dependence - 37Hierarchical Linear Modelling (Mixed Models & Multi Level Modelling) - purposetakes into account substructures, hidden subgroups, non independent data, greater flexibility and versatility (captures subgrouping variance - can capture interesting contextual questions)
- 38MLM - about levels and nestingmany levels to MLM and cases can be nested within the levels (lower levels require more cases than higher levels)
i.e. Level 1 = individual, Level 2 = class, Level 3 = year, Level 4 = school - 39Fixed versus Random effectsFixed - constant across individuals (i.e. age) (run by selecting all given known levels of variables and prediction based on variable)
Random - vary (test relationship between individual level predictors and dependent measure, varies depending on group they belong in - only sample some of possible levels of variable) - 40HLM and MLM problemsnature of relationship observed may differ depending on sub groupings
can lead to fallacy of composition (thinking what is true for 1 is true for all)
intercepts and function slopes may differ - 41Cross Sectional - purpose & processtests for hierarchical structures (HLM)
how much of the variance is explained by individual variations vs differences between groupings - 42Repeated Measures - purpose & processa type of longitudinal data analysis where repeated observations on individuals are considered data nested within individuals (RM data level 1; individuals level 2)
designed to avoid problems of covariance assumptions in traditional HLM RM - 43HLM/MLM - processwhen model building -2LLR is used to test for model improvements - test statistic gets smaller as model improves
- 44Variables in HLM/MLMSubject - variable ID
factors (gender)
covariates (BMI height metric measures)
repeated factors
combination vs interactions to test for variances in relationships
predictor group as random effect vs predictor and group as combination - 45Power analysis - purpose & what it can be used fortests for sample size (priori)
power levels (post hoc)
and effect size
can be used to determine the sample size required for research evaluate study results including effect size & power study had - 46Power analysis - how?can determine any of the following as long as you have the other 3
- power
- alpha
- sample size
- effect size - 47Power analysis - problemsuseless outside of NHST framework, T1 and T2 errors implications due to interpretation of NHST
- 48—
Information
- Description
- Flashcards on Stats Flashcards SEM1, created by Isabella Loxton on 19/06/2018.
- Created by
- Isabella Loxton
- Views
- 0
- Rating