Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
53 Cards in this Set
- Front
- Back
Coefficent of Determinations
|
R2 - how well the line of agression approximates your data
|
|
So if you reject your null cuz X2 is bigger then X critical.....
|
then you know your data isnt normally distributed.
So do ICC - reliability amoung raters |
|
what tests reliability amoung raters
|
ICC
|
|
what type of error effects reliability
|
random
|
|
is chi squared reliable?
what if you want to do a chi square but you dont have nutually exclussive data? |
no.
becuase it is just a comparison then do McNemar Test |
|
You did a nonparamentric matched test but think there is a problem with variance.
|
do a wilcoxcen T test
|
|
Power =
|
the probability to correctly reject the null
|
|
4 ways to discribe variability
|
variance
variation coefccient (%) standard deviation (units) Range (no outliers) |
|
R =
|
strength of correlation
|
|
What if you want to Do an ICC but only have ordinal data?
|
do Kappa
|
|
How do you check the internal consistancy of either a ICC or Kappa?
|
chronbachs alpha
|
|
How do you interpret a statistical ratio?
|
1 = different due to sampling error
+1 = treatment effect |
|
If t > t critical
if u < ucritical |
reject null
reject null |
|
if p < alpha...
|
reject null
|
|
How do you interpret correlation?
|
1.0 = perfect
0 = no correlation if p<alpha = significant correlation |
|
what if X2 > X critical
|
reject the null
|
|
effect size =
|
meaure of the strentgh of the relationships between 2 variables
|
|
paramtertic data
1 ID 2 levels IV Nonparametric data? |
unpaired t test
U test |
|
2 levles, 1 IV, paired test
parametric data = nonparametric data = ordinal data = |
paired t
sign test u test |
|
You want to do multiple t tests (1IV, 3 levels)
and you reject the null so you what if you realize uyou have unequal variance? |
ANOVA
run SR, then a pot hoc correct with epsilon ghg or hf |
|
Nonparametric data
correlated paired samples binomial info nonparametric data correlated paired samples ordinal data |
sign test
wilcoxen signed ranks |
|
non parametric data
2 groups correlated nonparametric data 2 group ID measures |
friedman 2 way
Kruskal wallis - then do man whit for post hoc and then a bonferroni correction |
|
with Sign test
x = n = |
x = # fewer signs
n = # differences |
|
T < T crit
T = |
reject null (wilcox)
sum of least frequent direction |
|
You did a nonparamentric matched test but think there is a problem with variance.
|
do a wilcoxcen T test
|
|
Power =
|
the probability to correctly reject the null
|
|
4 ways to discribe variability
|
variance
variation coefccient (%) standard deviation (units) Range (no outliers) |
|
R =
|
strength of correlation
|
|
What if you want to Do an ICC but only have ordinal data?
|
do Kappa
|
|
4 Strength of ICC
|
1. can asses reliability amoung two or more raters
2. doesnt require same numbers of raters for each subject 3. designed for interval and ratio data 4. can be used with nominal data |
|
If you have low power, what do you need to protect against a type two error?
|
more subjects
|
|
Name two non parametric tests for correlated paired samples
|
1. wilcox
2. sign test ranked |
|
What test would you use for MMT ordinal data (just memorize this)
|
sign
|
|
What does wilcoxen test take into account that sign test doesnt?
|
The direction and the magnitude of the differences. If the data is ordinal this test has more POWERRRRR
|
|
Name 2 nonparametric ANOVAS
|
1. Kruskal wallis One way analysis of variance of ranks
2. Friedman two-way analysis of variance by ranks |
|
If you had 2 levels of 1 independant variable and paired data....
1. and you had parametric data 2. And you had non parametric data (nominal ) 3. and you had nonparametric data (ordinal) |
1. paired t test
2. sign test 3. U test |
|
After your t test, if the data now represents a different population...
|
then your treatment worked
|
|
How do we know how big the difference has to be before we can be sure that its due to treatment? t test
|
statistical ratio
becuase if difference is due to sampling error then the statistical ratio will be 1. if the difference is due to treatment effect the statistical ratio will be greater then 1! |
|
variability within groups means-
|
this is your estimation of sample error. Stastical ratioes take this into account. This number is usually 1
|
|
name 4 assumptions for an independant sample t test
|
1. independant groups
2. RA 3. normal distrobution 4. variance of the two groups is relatively equal |
|
what 3 things do you need to be able to find t critical?
|
1. power
2. one or 2 tailed test 3. DOF |
|
So running a two tailed test would make it ____ to reject the null hypothesis since...
|
easier
since power is 2.5% insterad of 5% |
|
ANOVA means literally...
|
analysis of variance
|
|
just memorize this bullshit: for more than two groups we use __________ to represent the variablity
|
sum of squares
|
|
Standard error, generated from theortical populations, is used to find the
|
confidence interval: is a range of scores with speciic boundaries that probably contains the population mean
to inc confidence you INC the interval as sample size inc, the interval gets smaller |
|
____ = predetermined probability of making a type I error (alpha is chosen a priori)
|
level of significance
|
|
If your sample size was too small your descriptive test...
|
fails
|
|
coefficent of variation describes variability as a...
|
proportion of the mean
|
|
The two most common methods of data collection (I think for qual is..)
|
1. interviews
2. observation |
|
what type of research is NOT linear?
|
qualitative
|
|
correlation magnitude =
|
1.00 is perfect
0.00 is NO correlation 1.00- looks like all dots on the same line |
|
Which correlation to run..
1. ratio data 2. Ordinal data 3. Dichotomous Variable 4. 1 dichotomous, 2 continues |
1. (r) pearson product moment correlation coeff (measures how far each one is from a point
2. Spearman Rank correlation Coeff 3. PHI coeff 4. point diserial or rank biserial |
|
unspoken truth
|
if a group of scores used to evaluate the reliability of a measure is too HETER, you will get a good reliability coefficient, even if the errors are large
|