5 Statistics You Should Know Being Smart Reading Statistics Martha - - PowerPoint PPT Presentation

5 Statistics You Should Know

Being Smart Reading Statistics Martha J Lentz

Funding Support: SON Office for Nursing Research P30 NR04001

So What are the Stats

Effect size Plots/Descriptive Correlations Tests of Difference Time ordered

How do I know which one to use?

What is the question

Types of Questions

Look for action verb Describe Compare Association among Count the number of groups What kind of dependent variable

So What are the Stats

Effect size Plots/Descriptive Correlations Tests of Difference Time ordered

Effect Size

Lets us know was the sample big enough Mean1 – Mean2/ sd Do try this at home

Find These Effect Sizes d

Sample 12 women POMS Anger Scale mean1=2.25 mean2=1.44 sd=2.84 d=0.28 Bodily Feeling Muscle Pain mean1=17.17 mean2= 15.0 sd 1.46 d=1.49

Rule of Thumb Effect Sizes

Small = .2 Medium = .5 Large = .8

Sample Per Group Need for 80% Power

Effect Size .2 .3 .5 .6 .7 .8 1.0 1.40 Sample 393 175 64 45 33 26 19 10

Statistical Significance is not Clinical Significance

A non statistically significant effect may be clinically important

So What are the Stats

Effect size Plots/Descriptive Correlations Tests of Difference Time ordered

Plots

Plot your data Look at plots in papers, do they really look like what author is claiming.

Scatter Plot HT vs. WT

Line Plot eNo

Comparison

UB+8 hours UB+4 hrs usual bedtime (UB)

Mean eNO

30.00 25.00 20.00 15.00 10.00 5.00 UB+8 hours UB+4 hrs usual bedtime (UB)

Does the child have a diagnosis of asthma?" yes no

sleep delay night baseline night

Error Bars eNO Clustered by Group

Does the child have a diagnosis of asthma?"

yes no

95% CI Average eNO_2

35.00 30.00 25.00 20.00 15.00 10.00 5.00 sleep delay noc baseline noc

labprot

Descriptive

Mean Median SD

Look at sd Compared to Mean

Sample 12 women POMS Anger Scale mean1=2.25 mean2=1.44 sd=2.84 Bodily Feeling Muscle Pain mean1=17.17 mean2= 15.0 sd 1.46

So What are the Stats

Effect size Plots/Descriptive Correlations Tests of Difference Time ordered

Correlation

Bivariate- between two variables Regression- several variables to one

• utcome variable

Need variability

Scatter Plot HT vs. WT r=.72 r2=.52

Rule of Thumb

Little r < 0.4 is not really meaningful Square r and look at % variance explained <16% not really meaningful Regression look at change R2 apply above rule

So What are the Stats

Effect size Plots/Descriptive Correlations Tests of Difference Time ordered

Tests of Difference

How many independent groups Parametric test- estimate parameter e.g. a mean T-test two groups ANOVA multiple groups Non-Parametric Mann-Whitney U two groups Kruskal-Wallis multiple groups

Parametric or Non-Parametric

Use Parametric when Have appropriate distribution required to estimate a parameter such as mean Use Non-Parametric when Have a funky distribution Need to do calculation by hand

Good Distribution Use Parametric

Total score: Attitude Toward Women Scale

150.00 125.00 100.00 75.00 50.00

Frequency

80 60 40 20

Total score: Attitude Toward Women Scale

Mean =120.63

• Std. Dev. =17.189

N =652

Not So Good Use Non-Parametric

total bodily feel score noc 5

140.00 130.00 120.00 110.00 100.00

Frequency

5 4 3 2 1

Histogram

Mean =115.83

• Std. Dev. =8.211

N =12

Another Reason to Plot Your Data

So What are the Stats

Effect size Plots/Descriptive Correlations Tests of Difference Time ordered

Time Ordered Data

It is correlated It violates all assumption of independence of most stats It is very hard to get don’t waste it

Time Ordered Data Types of Tests

Paired T-Test 2 points in time Repeated Measure ANOVA multiple points in time More then 4 need special time series tests

Time Ordered Data Abuse

It is flat out wrong to treat each time point as a separate subject and use in a conventional statistical test Look at sample N if few in number and df or n listed for test is a lot it is wrong

Have Fun Reading Statistics