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;
48 Cards in this Set
- Front
- Back
binary or dichotomous variable
|
only two values. yes/no. male/female
|
|
ordinal variable
|
there's an inherent ranking. but they do not be scalar. ex. low, medium, or high prognosis. but unlike continuous variable, the different b/w low and medium rank do not have to be the same difference between medium and high rank.
|
|
if there was a perfectly normal curve then what 3 things should be same?
|
mean, median, mode
|
|
disadvantages of mean
|
1. doesn't tell distribution
2. affected by outliers |
|
advantage of median
|
less affected by outliers than mean
|
|
skewed distributions. positive skew means?
|
that the longer tail is on the positive end. so the mean> median.
|
|
nonparametric tests is used when the samples are?
|
when sample size is small or samples are noncontinuous.
nonparametric tests makes no assumptions about the underlying distributions of the data being compared in the 2 groups. |
|
deviance = ?
|
sum ((xi-X)^2)
|
|
variance = ?
|
deviance/(N-1)
|
|
STD = ?
|
sqr(variance).
X+/- 1.96 of STD = 95% confidence interval |
|
4 different ways to do bivariate comparisons?
|
1. absolute difference (15%-10%)
2. relative difference (15%/10%) 3. percent difference ((15%-10%)/15%)) 4. comparison of the means |
|
alpha
|
probability that represents the experimenter's tolerance for rejecing the null hypothesis due to chance alone.
when alpha is 0.05, then when null hypothesis is rejected there is only 5% chance that the data difference actually happened by chance. lower the alpha set, more rigorous the standard for testing the null hypothesis. |
|
p value
|
the probability that the difference in data happened by chance.
less the p value, less likely that the difference happened by chance reject null hypothesis when p<alpha accept null hypothesis when p>alpha |
|
chi square is used to ?
|
to compare the 2 distributions of scores. sum of (O-E)^2/(E)
to compare noncontinuous data in two groups. |
|
student's t test
|
parametric test to compare 2 means derived from 2 samples
|
|
paired t test
|
used when an observation is made twice on a sample.
t = Difference / (squroot (std error of difference/number of subjects)) |
|
examples of permutation tests
|
Fisher's exact test
Pittman Welch permutation test these consider all possible permutations of the data and compare them w/ the actual values |
|
ranking tests
|
Ex. Wilcoxon test, Mann-Whitney U test
|
|
Spearman rank order formula
|
used for data that is continuous and based upon some ranking. this is for making correlation coefficient
|
|
Pearson product-moment correlation formula
|
used for ranked data that have no intrinsic numerical value.
|
|
a positive correlation between "a" and "b" can be interpreted as ? (3 different ways)
|
1. a causes b
2. b causes a 3. c causes both a and b |
|
beta tests for ?
alpha tests for? (false ___) |
beta - false negative (type II error).
accepting null hypothesis when it is not true alpha - false positive (type I error). rejecting the nulll hypothesis when there is no significant difference |
|
Power
|
probability of rejecting the null hypothesis when it is false. higher the better.
power = 1 - beta. so you want smaller beta. higher the power, the more ability the study had to detect a true difference between the two groups when pB <0.05, there is less than 5% chance that the study was not powerful enough to detect a true difference. |
|
explained variance (R2)
|
higher the explained variance, the more likely it is that the independent variable is responsible for all of the variation.
|
|
in multilinear regression analysis, when the m is bigger
|
it means that that variable has a bigger effect
|
|
in logistic regression the y is
|
a binary variable
|
|
discriminant function analysis, the outcome of interest is __
|
categorical
ex. bilirubin AST, ALT when testing liver functions |
|
3 different types of multivariable models
|
1. logistic regression
2. ANOVA 3. discriminant function analysis |
|
sensitivity is?
|
probability that the individuals w/ the disease would be correctly identified as having the disease.
|
|
specificity is?
|
probability that the individuals w/o the disease would be correctly identified as nondisease by the test.
|
|
PPV is?
|
probability that the individuals who test positive for the disease actually have the disease.
|
|
NPV is?
|
probability that the individuals who tested negative for the disease actually do not have the disease.
|
|
advantages of randomized controlled trial?
|
it is considered to be the most rigorous and powerful approach to answering clinical q's.
|
|
cross-over randomized controlled trial: adv? when can it not be used?
|
adv: this adds analytical power over the randomized trial. this can decrease the number of samples
not used: when disease change very rapidly or are unpredictable |
|
non-experimental designs are used when?
|
when the disease is rare and it would be impossible to follow 1000s of samples for years. so you do a retrospective study
this will create more bias |
|
4 types of nonexperimental designs
|
1. case-control
2. cohort 3. cross sectional 4. longitudinal |
|
case control study is?
|
identify ppls on the basis of the same outcome/disease and previous experiences w/ various exposures.
case = ppls w/ cancer control = w/o cancer cases and controls are the same except that they have cancer or not. |
|
odd ratio
|
used in case-control study. ad/bc.
if the odds ratio is 7.5 and then pt w/ cancer is 7.5 times more likely to have been a smoker. |
|
cohort study is?
|
pts w/ common experience (like exposure) are assembled and monitored until an outcome appears.
use relative risk and attributable risk. relative r = incidence (exposed) over incidence unexposed. attributable risk = incidence exposed - incidence unexposed |
|
advantage of cohort study over case-control studies
|
1. less bias, like recall bias
2. can control level of exposure/experience cohort = definitive observational clinical study |
|
crude birth rate is?
|
# newbirth/total population. can be misleading for an indicator of pop growth cuz it doesn't show # of baby death
|
|
crude mortality rate is?
crude rate of natural increase |
# of deaths from all causes per year divided by the total population
crude birth rate - crude death rate |
|
infant mortality rate?
neonatal mortality rate postnatal mortality rate |
# of deaths until 1st yr/ total number of live births
w/in first 4 wks after 4 wks but before 1st bday |
|
life expectancy is?
|
median survival time.
uses cross-sectional mortality rates for each age to summarize mortality across the life-cycle for a given population. |
|
detection bias
|
more info is solicited from teh tx group asopposed to the placebo group. solved by double-blind studies
|
|
proficiency bias
referral bias (sampling bias) |
the intervention under study is delivered w/ unusual skill or incompetence so can't be reproduced
sampling error |
|
susceptibility bias
recall bias ascertainment bias |
when pts w/ severe conditions get a different intervention. do "randomizing" of patients
pts can't remember well, this is in retrospective studies pts w/ suspected outcome are more extensively probed. |
|
lead time bias
measurement bias |
when screening test can detect a disease earlier than before, and ppls think that that test is allowing longer survival of the pts
the act/fact of measuring changes what is being measured (ex. BP) (ex. Hawthorne Effect = when pts who know they are being assessed act differently becasue they are being observed) |