Statistics In Heath Sciences Final

Post Hoc Tests

tell us which groups differ from the rest

Scheffe Test

• post hoc test

• comparing all pairs of two means when the sample sizes are unequal

• equal or unequal sample sizes

Tukey Test

• post hoc test

• comparing all pairs of two means when the sample sizes are equal

• equal sample sizes

What is the symbol for the Tukey test?

q

Bonferroni Test

• post hoc test

• probably the simplest but most conservative

• least likely to make a type 1 error

What type of data is used for a histogram?

used to visualize the distribution of continuous, numerical, or quantitative data

Main difference between standard deviation graph, positively skewed graph, and negatively skewed graph?

the shape of the graph

Standard deviation graph

• normal distribution

• symmetrical shape: mean, median, mode are all the same

• data is spread evenly on both sides

positively skewed graph

• skewed to the right

• tail stretches to the right

• most data is on the left

example of a positively skewed graph

income distribution (a few people earn very high incomes)

Negatively skewed graph

• skewed to the left

• tail stretches to the left

• most data is on the right

example of negatively skewed graph

easy test scores (most people score high, few score low)

What is the empirical rule?

• 68-95-99.7

• tells you how the data is being spread around the mean

• normal distributions (bell-shaped curve)

How much data falls within one standard deviation of the mean?

68%

How much data falls within two standard deviations of the mean?

95%

How much data falls within 3 standard deviations of the mean?

99.7%

What is the symbol of correlation coefficient?

r

What does the correlation coefficient tell you?

1. direction (positive or negative)

2. strength (how strong the relationship is)

r value of a positive correlation

r>0 as one variable increases, the other increases

r value of negative correlation

r<0 as one variable increases, the other decreases

what does an r value of 0 mean?

no correlation

for strength, what does an r value close to 1 mean?

strong correlation

does correlation mean causation?

NO

definition of statistical power?

probability that a test correctly detects a real effect (rejects the null hypothesis)

if a study has 80% power what does that mean?

80% chance you’ll detect a true effect and a 20% chance you’ll miss it

when to not use a parametric test

• data is not normally distributed (skewed)

• small sample size

• data is ordinal (ranked)

• extreme outliers

• data is categorical

what nonparametric test should you use when comparing two independent groups?

mann-whitney U test

what nonparametric test should you use when comparing two paired groups?

wilcoxon sign-rank test

what nonparametric test should you use when comparing three plus independent groups?

kruskal-wallis test

what nonparametric test should you use when comparing three plus paired groups?

friedman test

what nonparametric test should you use when looking at relationships between groups?

spearman rank correlation

what are nonparametric tests?

they use rank, are more flexible, and slightly less powerful

odds and risk ratio of 1

no difference between groups

odds and risk ratio >1

increased risk/odds

odds and risk ratio <1

decreased risk/odds

if a 95% confidence interval includes 1…

its not statsically significant

if a 95% confidence interval does NOT include 1…

its statistically significant

relative risk of 1.5 and confidence interval of 1.2-1.8

• 50% higher risk

• statically significant

selection bias

groups being compared are not truly similar at baseline (healthier people are more likely to enroll in study)

sampling bias

your sample does not represent the population (only surveying college students for general population health)

recall bias

people don’t accurately remember past events (sick patients remember exposures more than healthy controls)

observer bias

researcher’s exceptions influence results (researchers subconsciously rate treatment groups better)

interviewer bias

asking questions differently between groups

response bias

participants give inaccurate answers (underreporting alcohol use)

attrition bias

people drop out and it affects results (sicker patients leave the study)

confounding bias

a third variable distorts the relationship (coffee appears linked to lung cancer, but smoking is the real cause)

publication bias

positive results are more likely to be published

lead-time bias

early detection makes survival time look longer without changing outcome

length-time bias

screening detects slow, less aggressive disease

categorical (qualitative) data

data that falls into groups or categories

examples of categorical data

blood type, gender, yes/no outcomes, injury type

continuous (quanitative) data

data that is numerical and measurable, answers how much or how many

examples of continuous data

height, weight, BP, time

nominal data

in name only, most primitive, categorical and qualitative, cannot add, subtract, multiply divide

examples of nominal data

heart rate, nm wavelength, height, weight, cost, distance, age, salaries

advantages of a cross over study

• each person is their own control which (reduces differences)

• small sample size

• more statical power

• efficient comparison of treatments

disadvantages of cross over study

• carryover effects

• not suitable for all conditions

• washout periods could be issue

• time related effects

• dropouts

what is the intention to treat analysis?

when you analyze participants in the groups they were originally assigned to, regardless of what actually happened later

how to use intention to treat analysis?

if someone stops treatment, switches groups, or doesn’t follow protocol, you still have to analyze them as if they stayed in their original group

number to treat equation

• 1 patient to benefit

• 1/control event rate-experimental event rate

number to harm

• 1 patient harmed

• 1/experimental event rate- control event rate

why would you want to do a meta-analysis?

to combine results from multiple studies addressing the same question to get a stronger and more reliable conclusion

reasons to do a meta-analysis

• increases statistical power

• improves precision

• resolves conflicting results

• generalizability

• identifies patterns or subgroups

• foundation for evidence-based medicine

sensitivity

measures a test’s ability to correctly identify people who have the disease

specificity

measures a test’s ability to correctly identify people who do NOT have the disease

when do you use a randomized control trial?

• studying a treatment, medication, or intervention

• its ethical and feasible

• you want the strongest evidence for efficacy

when do you use a case control study?

• disease is rare

• disease takes a long time to develop

• you want to study possible risk factors

when to use a retrospective cohort study?

• exposure is rare

• data already exists

• randomization would be unethical

requirements for data to be parametric

• normal distribution

• continuous data

• equal variance

• independence of observations

• absence of outliers

what does the P in PICOT stand for?

participants/population

what does the I in PICOT stand for?

intervention/indication

what does the C in PICOT stand for?

comparator/control

what does the O in PICOT stand for?

outcome

what does the T in PICOT stand for?

time frame

in studies with very small numbers, what is the best type of test to use categorical or continuous?

categorical (nonparametric) because they make fewer assumptions about the data distribution

what is ratio data?

data that has a true zero

r value of -0.66 would represent what?

a strong negative correlation

mean

average

median

middle number

mode

most frequency occurring number in data set

what is different about the variance equation for population vs sample

• population: divide by N

• sample: divide by N-1

how to find the standard deviation?

square root of variance

when to use z score?

population and you know standard deviation

when to use t score?

standard deviation is unknown and small sample

How to find E for chi squared?

population percentage multiplied by sample size

z score for 90% CI

1.65

z score for 95% CI

1.96

z score for 99% CI

2.58