Handout16 - Lecture 16 1. Risk and odds 2. Smoothing binary...

Info iconThis preview shows pages 1–6. Sign up to view the full content.

View Full Document Right Arrow Icon
Lecture 16 1. Risk and odds 2. Smoothing binary data 3. Logistic regression with binary response 4. Proc Logistic surprises Logistic regression using the SAS system: theory and application by Paul Allison, 2001, Wiley-SAS Categorical Data Analysis Using the SAS System, 2nd ed. by Stokes, Davis, and Koch, 2009, SAS Institute “Proc Logistic: Traps for the Unwary,” by P.L. Flom (on course website) 1 2 £ 2 Tables: Relative Risk, Odds Ratio group event Frequency| Row Pct | 0| 1| Total ---------+--------+--------+ 1| 141| 33| 174 |8 1 . 0 3| 18.97 | ---------+--------+--------+ 2| 187| 84| 271 | 69.00 | 31.00 | ---------+--------+--------+ Total 328 117 445 row percent = rate of events = risk of events Comparisons: • risk difference • risk ratio (relative risk) • odds ratio 2
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Two different null hypotheses for risks 1. H 0 : p a ° p c = 0, risk difference is zero, Z -test, based on ( ˆ p a ° ˆ p c ), is equivalent to chi-square test Alternative test of risk differences when some cells have small counts: Fisher’s exact test. 2. Risk ratio , or relative risk =1 : H 0 : p a p c = 1 Null value for differences is 0. Null value for ratios is 1. 3 Odds odds = number of events number without event in sample Event No Event odds risk Group 1 A B A / BA /( A + B ) Group 2 C D C / DC /( C + D ) Relating odds to risk: odds = number with event number without event = number with event ± n number without event ± n = ˆ p 1 ° ˆ p For rare events, ˆ p º 0 and so the denominator is almost 1, and odds º risk. 4
Background image of page 2
Event No Event odds risk Group 1 A B A / BA /( A + B ) Group 2 C D C / DC /( C + D ) Odds are compared only by ratio, never by difference. Odds ratio is the odds in the top row divided by odds in the bottom row, which simpliFes to AD ± BC . Test whether population odds ratio is one, H 0 : OR = 1 by checking whether the 95% conFdence interval covers 1. 5 Comparing risks and odds in Proc Freq 6
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
group event Frequency| Row Pct | 0| 1| Total ---------+--------+--------+ 1| 141| 33| 1 7 4 | 81.03 | 18.97 | ---------+--------+--------+ 2| 187| 84| 2 7 1 | 69.00 | 31.00 | ---------+--------+--------+ Total 328 117 445 7 Statistic DF Value Prob ------------------------------------------------------ Chi-Square 1 7.9142 0.0049 Fisher’s Exact Test ---------------------------------- Cell (1,1) Frequency (F) 141 Left-sided Pr <= F 0.9985 Right-sided Pr >= F 0.0031 Table Probability (P) 0.0016 Two-sided Pr <= P 0.0057 Estimates of the Relative Risk (Row1/Row2) Type of Study Value 95% Confidence Limits ----------------------------------------------------------------- Case-Control ( Odds Ratio ) 1.9193 1.2138 3.0348 Cohort (Col1 Risk) 1.1743 1.0548 1.3075 Cohort (Col2 Risk) 0.6119 0.4291 0.8725 Sample Size = 445 8
Background image of page 4
Binary responses Event ± no-event, or 0 ± 1 responses are binary responses . Set-up: one trial in which Y takes values 1=event ± 0=no event. Chance of event = P [ Y = 1] = º = population event rate Chance of no event = P [ Y = 0] = 1 ° º Y has Bernoulli distribution (after Jakob Bernoulli, 1654–1705). mean of Y = º , standard deviation of Y = p º (1 ° º ). SD is a function of the mean, unlike Normal distribution. 9 Binary data comes as individual responses (0 or 1) and also as grouped responses: Grouped binary responses: If many trials under done with same chance of event, such as number of heads in n coin Fips, then results are grouped: X = k events in n trials.
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 6
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 11/21/2011 for the course PUBH 6470 taught by Professor Williamthomas during the Fall '11 term at University of Florida.

Page1 / 25

Handout16 - Lecture 16 1. Risk and odds 2. Smoothing binary...

This preview shows document pages 1 - 6. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online