Flashcards
General MCQ Statistics 2
Cards in this deck (55)
- 1what is an unpaired design?a between-subject design comparisons are made between values recorded from DIFFERENT INDIVIDUALS
applied to two independent designs
eg. diabetic patients and non-diabetic patients - 2what is a paired design?in a within-subject design, comparisons are made between values recorded from the SAME INDIVIDUAL
data is derived from study subjects who have been measured at TWO TIME POINTS (so each individual has two measurements)
eg. the two measurements are generally before and after a treatment intervention
--> same people before and after - 3in a paired design, why is the treatment ALWAYS counter-balanced and random?to control for potential confounding effect of order
- 4when would we use a paired t-test?to test the means of the differences between 2 groups of paired data
- 5how would you find the t value of a paired t-test?(sample mean difference - hypothesised mean difference)/ standard error of the mean difference
- 6what is an assumption of the paired t-test?that the individual DIFFERENCES are normally distributed
DIFFERENCES should be independent of one another - 7how must paired data be presented?all data points for each group plotted and lines drawn between them (see notes)
- 8if the normal distribution curve of each group have similar means and SDs overlap then we become uncertain to whether we have two separate means, what do we use?unpaired t-test
- 9how do you work out degrees of freedom in an unpaired t-test?(n1 + n2 - 2 )
- 10how do you work out the t value of an unpaired t-testt = (difference in means - 0)/ standard error of difference
0 - because Ho states that there is no difference - 11what are the 3 assumptions of an unpaired t-test?1) data from normally distributed population
2) statistical independence - no pseudo replication
3) homogeneity of variance --> similar variance in each group - 12when/why would we use a Levene's test?to see if the standard deviations are similar or different (p< 0.05)
TEST FOR HOMOGENEITY OF VARIANCE - 13what is the difference between unpaired and paired t -tests in terms of sample sizepaired HAS to have same number in each condition that you are pairing
unpaired doesn't have to have same n in each sample group - 14what can we conclude if the confidence intervals for an unpaired t-test span 0?we can confidently say that the means are not significantly different
difference between the means COULD be 0 - 15what are confounding factors?factors that are correlated with the independent variable being manipulated and affect the independent variable being measured
eg. no drownings correlate with the number of ice cream sales
rise in temperature = CONFOUNDING VARIABLE
prevents us from drawing the correct conclusions from our study
good experimental design has to control all likely confounding factors - 16what is a double blind trials and what does this avoid?neither the researchers or patients know what they are getting
- 17when do we use a one-way ANOVA? single factor analysis of variancewhen you need to simultaneously compare the differences between more than 3 groups of data whilst remaining in control of type 1 error rate of 0.05 overall
DIFFERENCE IN MEANS OF TWO OR MORE UN-MATCHED GROUPS - 18does an ANOVA need a factor?yes
eg. eye colour - brown, green, blue - 19in an ANOVA what does an F ratio of less than 1 imply--> more variation within the individual groups than between them
- 20what are the 3 assumptions of an ANOVA test?1) normal distribution of the residuals (and data) in each group
2) statistical independence of each data point
3) homogeneity of variance (SD or variance similar) --> LEVENE'S TEST - 21if the assumptions for ANOVA are not met, then what should you do?transform the data and if not use Kruskal Wallis test
- 22what is the non-parametric equivalent of ANOVA?kruskal-wallis
- 23what do we mean by a balanced design? is this more powerful?equal numbers in each group
yes - 24how do you reach a conclusion using an ANOVA?get a p-value and then reject or accept the null hypothesis
- 25if the null hypothesis is REJECTED in the ANOVA, what do we do?conduct a post hoc test
to determine which means are likely to differ - 26what is the multiple comparisons post hoc test we must choose EVERY TIME for the exam?bonferroni
- 27why is Bonferroni said to be a conservative post hoc test? what type of error is it good at preventing and what error is it most likely to give?good way to prevent type 1 error (fans positive)
may give away false negative type 2 errors more readily as a trade off - 28increasing the number of comparisons you are doing does what to the power of the test?reduces it
- 29name the two multiple comparison tests?1) bonferroni
2) tukey - 30when is the Tukey test good?good if sample size and variances are equal
- 31if asked to conduct a MULTIPLE CONTRASTS test, what would be the choice?dunnet's
multiple comparison test that looks at all groups vs control - 32what is the problem with most post hoc tests?most are robust to small deviations from normality
if the sample groups are unequal and variance is large --> not useful - 33give two categorical (qualitative) data typesnominal
ordinal - 34what is nominal data?collection of categories that are mutually exclusive but not ordered
eg. sex / blood group - 35what is ordinal data?like nominal but with a rank order to the categories
eg. 3rd, low 2,2, high 2,2 - 36give 2 numerical (quantitative) types of datainterval scale
ratio scale - 37what is the interval scale?rank order has consistent separation between values following addition and subtraction
eg. date - 38what is the ratio scale?like interval scale data but with an absolute zero so that multiplication and division become meaningful
either CONTINUOUS scale (mass, length, duration)
or DISCRETE (no. eggs laid, no. labelled cells counted) - 39how can we compare the differences in ordinal data?non - parametric test
- 40when do we use non-parametric tests?when the assumptions of the parameter of the test have not been met
THEY DO NOT RELY ON THE DATA HAVING A NORMAL DISTRIBUTION
are based on RANKING the data
very good for testing non-numerical data and when we can't meet the assumptions of parametric tests - 41what are the 2 assumptions of non-parametric tests?1) randomly sampled data
2) statistical independence - 42what is the non-parametric version of the paired t-test?wilcoxon signed rank test
- 43what is the non-parametric version of the unpaired t-test?mann-whitney u
- 44non-parametric version of one-way ANOVA?kruskal-wallis test
- 45how is a wilcoxon-signed rank test carried out?1) create a column that calculates the differences
2) create a column that then ranks these differences
3) create a signed rank column (if the difference is negative then the rank is also e.g. diff = -1.9 rank = -2)
4) add the negative ranks up and then add up the positive ranks and then add up
total positive ranks = total negative ranks
any difference between positive and negative ranks is looked up in a wilcoxon table and give associated p -value - 46how do you decide on sample size?1) look at similar earlier experiments
2) perform pilot study
3) conduct a formal power analysis - 47what is statistical power?the probability that you will be able to detect a difference if one is really there to be detected
- 48how is power decreased and increased?INCREASED
effect size high
sample size high
DECREASED
random variation low - 49what is the notation for effect size?r
- 50what is the equation to calculate power? involves type 2 error (beta)power = 1 - type 2 error rate (beta)
- 51what size of effect is r = 0.1?small effect and explains 1% of the total variance (1squared)
- 52r = 0.3medium effect and explains 9% total variance (3squared)
- 53r = 0.5high effect explains 25% of total variance
- 54ANOVA EQUATIONS?
- 55ANOVA, rsquared =?sum of squares (group)/ sum of squares (total)
square root this to get effect size, r
Information
- Description
- 2nd Year Statistics Flashcards on General MCQ Statistics 2, created by William Chapman on 11/05/2016.
- Created by
- William Chapman
- Views
- 12
- Rating