Lecture 4 (stats) Flashcards
1
Q
What are the steps of the empirical cycle?
A
- observation: the idea for the hypothesis
- induction: hypothesis, general rule
- deduction: prediction and operationalization
- testing: test the hypothesis and compare data to prediction
- evaluation: interpret results in terms of hypothesis
2
Q
theory?
A
set of principles explaining a general phenomenon
3
Q
hypothesis?
A
- explanation for a phenomenon which is informed and based on a theory
- predictions are derived from the hypothesis
4
Q
falsification?
A
disproving a hypothesis/theory
5
Q
independent variables?
A
cause, manipulated variable, predictor variable
6
Q
dependent variable?
A
outcome
7
Q
categorical variables?
A
- contain categories
- binary variable: if two options are available
- nominal variable: used to denote categories without an order
- ordinal variable: used to denote categories with an order
8
Q
continous variables?
A
- gives score on a scale and can take on any value of the scale used
- interval variables: need equal distances between the individual values
- ratio variables: require meaningful ratios of values in addition to equal steps between values (i.e. rating 4 is twice as good as rating 2)
- truly continuous variables can take on any value on the scale
- discrete variables usually only take on certain values
9
Q
measurement error?
A
- difference between actual true score and measured score
- can be due to usage of different measurement methods
10
Q
validity?
A
- does an instrument measure what its suppose to measure
- criterion validity: does an instrument measure what it is supposed to as established by certain criteria
- concurrent criterion validity: checking data using the new instrument and criteria for validity
- predictive criterion validity: if data can be used to predict observations at a later point in time
11
Q
reliability?
A
- does an instrument give consistent values for interpretation
- test retest reliability
12
Q
correlational research methods?
A
- involves observing natural events
- longitudional or cross sectional
13
Q
experimental research methods?
A
- introduce and take away an effect to establish causality
- confounding variable: hidden third variable that might be causing the cause effect link
14
Q
testing different entities?
A
- between-groups design: comparing results of different groups
- between-subjects design: each subject experiences only one condition
- independent design: no participant overlap between groups
15
Q
Manipulating the independent variable with the same entities?
A
- within subject design: type of repeated measures design where participants experience every condition
- repeated measures design: can be within subject design or pre and post intervention repeated measurements
16
Q
variation?
A
- unsystematic variation: small differences in measurement across conditions regardless of manipulation
- systematic variation: differences in performance in conditions due to manipulation
- in independent designs variation can be due to manipulation or due to differences on characteristics of the entities
- Randomization helps keep unsystematic variation to a minimum
17
Q
Systematic variation in repeated-measures designs?
A
- practice effects: different performance because of familiarity
- boredom effects: different performance because of boredom
- random assignment of the order of conditions helps eliminate this
18
Q
skeweness?
A
- lack of symmetry
- positively skewed: tail points to positive end and vice versa
19
Q
kurtosis?
A
- pointyness
- degree to which scores cluster at the ends of the distribution
- leptokurtic: positive kurtosis, lots of scores in the tails
- platykurtic: negative kurtosis, barely any scores in the tails
20
Q
frequency distribution?
A
plots how often data occur
21
Q
normal distribution?
A
- has a bell shape curve and is symmetrical
- kurtosis and skew are 0
22
Q
the mode?
A
- most frequent score
- graphs can be bimodal or multimodal if they have multiple modes
23
Q
the median?
A
- middle score of all scores when they are ordered according to magnitude
- when the data contains an even number of scores, the median is the average of the middle two values
- is unaffected by skew and extreme scores
24
Q
the mean?
A
- measure of central tendency, average
- Can be influenced by extreme scores
- Uses every score in the sample and is stable in different samples
25
Q
range of scores?
A
- dispersion, subtract lowest from largest score
- Affected by extreme scores
- solution: interquartile range which can be calculated by subtracting the top half median from the bottom half median
interquartile range is not affected by extreme scores
26
Q
deviance?
A
- can be calculated as deviance = X - mean of X
- for the total deviance you add up all deviance scores
27
Q
sum of squarred errors?
A
- indicates the total dispersion/error from the mean
- calculated as SS = sum of squared deviances
28
Q
standard deviation formula?
A
- s = the square root of (SS devided by N -1) with N being the total number of observations
- variance is s squared which represents the average dispersion
- (N - 1) represent the degrees of freedom, which signify the number of observations that are free to vary
29
Q
probability density functions?
A
- common probability distributions that can be used to calculate probabilities
- the area under the curve reveals the probability of certain events happening
- normal distribution with sd = 1 and mean = 0 most often used as data sets can be converted into this distribution
- z score calculation: (X - mean of X) divided by s
30
Q
reporting data?
A
- Scientific information about one’s findings should be shared openly and in much detail
- APA guidelines should be checked for correct reporting
- Guidelines exist on what notation should be used to represent statistics
31
Q
model fit?
A
how well a model represents the observed data
32
Q
linear and non linear models?
A
- linear: use a straight line to represent data
- non linear: curve the line to represent the data, can sometimes be more fit to represent the data but are also more complex
33
Q
how to predict the outcome?
A
- using the regression coefficient and a variable
- outcome of X = model + error of X
34
Q
how to calculate deviance?
A
- deviance symbolizes error
- deviance = outcome of X - model of X
35
Q
assessing the fit of a model?
A
- with the sum of squared deviances/erros (SS)
- for estimating a population parameter use variance formula
36
Q
sampling distribution?
A
- uses a large number of hypothetical samples to estimate the population parameters
- can reveal how representative a sample is of the population
37
Q
standard error?
A
- standard deviation of the sampling distribution
- reveals how widely the sample data are spread around the population parameter
- SE = standard deviation devided by squared root of N
38
Q
central limit theory?
A
with larger samples, the sampling distribution will approximate a normal distribution with mean and sd close to the population parameters
39
Q
confidence intervals?
A
- boundaries that are supposed to contain the true value of the population parameter for a percentage of the sample
- wider confidence intervals are worse representations of the true parameter
40
Q
how to calculate confidence interval?
A
- calculate z-score = (X - mean of X) divided by standard deviation
- bounds calculated by = mean of X +/- (z score x standard error)
41
Q
confidence intervals for smaller samples?
A
For n < 30 t-distributions can be used with the corresponding df = n - 1
42
Q
overlapping confidence intervals?
A
- help narrow down the range of plausible scores
- significantly different estimates: if 2 CIs do not overlap they most likely come from different populations
43
Q
p value?
A
p = 0.05 is used as a threshold for confidence because we want to reduce the probability of getting the results by chance alone
44
Q
types of hypothesis?
A
- H0: null hypothesis, no effect
- H1: alternative hypothesis, effect present
- accepting one hypotheis means that data is very likely under that hypothesis
45
Q
what are the steps for null hypothesis statistical testing (NHST)?
A
- Establishing hypotheses
- Establishing alpha, the significance level (usually 0.05)
- Establishing power (sample size needed)
- Calculate p-value and t test
- Compare p to alpha
- if p below or equal to alpha we have reason to reject H0
46
Q
one tailed test?
A
- aternative hypothesis says there is an effect in a specific direction (e.g., the mean is greater than or less than the specific value)
- smaller test statistic needed for significant result BUT only detects change in one direction
47
Q
two tailed tests?
A
- alternative hypothesis is different than 0, there is an effect in either direction
- larger test statistic needed for significant result
48
Q
type I error?
A
- rejecting the null when it is more likely to be true
- we believe there to be an effect but there is not one
- denotated by alpha and is the same as significance level which is equal to 1 minus confidence level
49
Q
type II error?
A
- accepting the null when we should reject it
- we believe there to be no effect but there is one
- when type II error increases type I error decreases and vice versa
50
Q
is it considered more harmful making a type I or a type II error?
A
- type I error since that means that science does not move foward
- context dependent (in medicine type II might be more harmful)
51
Q
bonferroni correction?
A
- If multiple tests are conducted, the type I error rate has to be adjusted (control for familywise error rate)
- formula = type I error divided by k, with k being the number of comparisons
52
Q
statistical power?
A
- probability that a test will find an effect if one exists
- depends on effect size, how large alpha (significance level) is, and sample size
