# Clinical Research Training Program 2021

0
0
1659 days ago, 729 views
PowerPoint PPT Presentation
2. Plot. Motivation behind Regression DiagnosticsResidualsOrdinary residuals, institutionalized residuals, studentized residuals, Jackknife residualsLeverage pointsDiagonal components of the cap lattice Influential observationsCook\'s separation CollinearityAlternate Strategies of Analysis. . 3. The procedures of relapse diagnostics are utilized to check the presumptions and to evaluate the precision of computa

### Presentation Transcript

Slide 1

﻿Clinical Research Training Program 2021 REGRESSION DIAGNOSTICS I Fall 2004 www.edc.gsph.pitt.edu/workforce/evade/clres2021.html

Slide 2

OUTLINE Purpose of Regression Diagnostics Residuals Ordinary residuals, institutionalized residuals, studentized residuals, Jackknife residuals Leverage focuses Diagonal components of the cap network Influential perceptions Cook's separation Collinearity Alternate Strategies of Analysis

Slide 3

Purpose of Regression Diagnostics The systems of relapse diagnostics are utilized to check the presumptions and to evaluate the precision of calculations for a relapse investigation.

Slide 4

MODEL ASSUMPTIONS Independence: the mistakes related with one perception are not associated with the blunders of some other perception Linearity: the relationship between the indicators X and the result Y ought to be straight Homoscedasticity: the blunder change ought to be steady Normality: the mistakes ought to be regularly dispersed Model Specification: the model ought to be appropriately determined (counting every single important variable, and barring unessential factors).

Slide 5

UNUSUAL OBSERVATIONS Outliers: In straight relapse, an exception is a perception with extensive leftover . At the end of the day, it is a perception whose needy variable esteem is uncommon given its qualities on the indicator factors X. An exception may demonstrate an example idiosyncrasy or may show an information section blunder or other issue.

Slide 6

UNUSUAL OBSERVATIONS Leverage: A perception with an extraordinary incentive on an indicator variable X is known as a point with high use. Use is a measure of how far an autonomous variable goes amiss from its mean. These use focuses can have a strangely vast impact on the gauge of relapse coefficients.

Slide 7

UNUSUAL OBSERVATIONS Influence: A perception is said to be persuasive if evacuating the perception significantly changes the gauge of relapse coefficients . Impact can be considered as the result of use and outlierness.

Slide 8

SIMPLE APPROACHES Detect blunders in the information & pinpoint potential infringement of the presumptions: Check the kind of subject Check the methodology for information gathering Check the unit of estimation for every variable Check the conceivable scope of qualities and a common incentive for every variable Descriptive insights

Slide 10

SIMPLE APPROACHES Analysis of residuals and other relapse indicative strategies give the most refined and exact assessment of model suspicions.

Slide 11

(surreptitiously) mistake term for the i th reaction RESIDUAL ANALYSIS Model i = 1, … , n Fitted model Ordinary residuals Difference b/w the watched and the normal results

Slide 12

• LEAST-SQUARES METHOD Birthweights (g/100) Estriol (mg/24 hr) levels of pregnant ladies

Slide 13

ORDINARY RESIDUALS The conventional residuals { e i } mirror the measure of inconsistency amongst watched and anticipated qualities that remaining parts after the information have been fitted by the slightest squares display. Fundamental supposition for in secret mistakes: Each lingering e i speaks to a gauge of the relating surreptitiously blunder  i .

Slide 14

ORDINARY RESIDUALS The mean of { e i } is 0. The gauge of populace difference registered from n residuals is s 2 is an unprejudiced estimator of  2 if the model is right.

Slide 15

ORDINARY RESIDUALS Ordinary residuals are associated (residuals total up to 0) and have unequal differences despite the fact that hidden mistakes are free and have level with change where h ii is the i th askew component of the ( n  n ) network H = X ( X " X ) - 1 X " , the cap lattice . Note (expected y = H · watched y ). { e i } are not autonomous irregular factors (aggregate up to zero). Nonetheless, if n >> p , then this reliance can be overlooked in any investigation of the residuals.

Slide 16

STANDARDIZED RESIDUALS Standardized lingering is characterized as The institutionalized residuals aggregate to 0 and consequently are not free. The institutionalized residuals have unit difference.

Slide 17

STUDENTIZED RESIDUALS Studentized lingering is characterized as where h ii is the i th corner to corner component of the ( n  n ) grid H = X ( X " X ) - 1 X " , the cap network . Note (expected y = H · watched y ). Esteem h ii ranges from 0 to 1, which is a measure of use. A high esteem implies more use, i.e., X is further far from the X-centroid (X-variable means).

Slide 18

STUDENTIZED RESIDUALS If information point is related with a higher use esteem ( h ii is bigger), we will get a greater studentized leftover esteem. Subsequently, residuals on the edging will have higher studentized remaining qualities (making them less demanding to single out). In the event that the information take after the typical suspicions for straight relapse, the studentized lingering around takes after t n - p - 1 .

Slide 19

LEVERAGE MEASURES The amount h ii , the use, measures the separation of the i th perception from the arrangement of x - variable means – in particular, from h ii shows that for a settled x i , when y i moves a tiny bit, what amount does move? In the event that moves a great deal, then y i can possibly drive the relapse, so the fact of the matter is a use point. Be that as it may, if barely moves by any means, then y i has no way of driving the relapse.

Slide 20

LEVERAGE MEASURES Under the model, we have Consequently, the normal use esteem is Hoaglin and Welsch (1978) prescribed investigating any perception for which h ii > 2( p +1)/n .

Slide 21

JACKKNIFE RESIDUALS Jackknife remaining is characterized as where is the MSE processed with the i th perception erased. On the off chance that the i th perception lies a long way from whatever remains of the information, s (- i ) will have a tendency to be considerably littler than s , which thus will make r (- i ) bigger in contrast with r i . On the off chance that the i th perception has bigger use esteem h ii , then r (- i ) will get to be distinctly bigger in contrast with r i .

Slide 22

JACKKNIFE RESIDUALS Therefore, residuals on the edging (high use) or far from whatever is left of the information (little S (- i) esteem) will have higher Jackknife lingering values (making them less demanding to single out). On the off chance that the standard suspicions are met, each folding blade remaining precisely takes after a t appropriation with ( n - p - 1) degrees of opportunity.

Slide 23

RESIDUALS

Slide 24

Graphical Analysis of Residuals Stem-and-leaf outline, histogram, boxplot, and ordinary likelihood plot of residuals can be utilized to test the ordinariness supposition. Studentized leftover (or pocketknife remaining) versus can be utilized to distinguish anomalies . Studentized remaining (or pocketknife lingering) versus can be utilized to distinguish nonlinearity . Studentized lingering (or folding blade leftover) versus (on the other hand x ) can be utilized to distinguish heteroscedasticity .

Slide 25

Graphical Analysis of Residuals Stem-and-leaf chart, histogram, boxplot, and typical likelihood plot of residuals can be utilized to test the ordinariness presumption. For the typical likelihood plot of residuals, if the information focuses are circulated far from the perfect 45 degree line, the ordinariness supposition is sketchy.

Slide 26

• relapse sbp age • foresee rstudent, rstudent • qnorm rstudent, network

Slide 27

Graphical Analysis of Residuals Studentized lingering (or folding blade leftover) versus can be utilized to identify exceptions . For an example with size sufficiently vast, 95% of Jackknife residuals ought to lie between + 2. For a specimen with size sufficiently expansive, 99% of Jackknife residuals ought to lie between + 2.5. Any perception for which the supreme estimation of the Jackknife residuals is at least 3 , is probably going to be an exception.

Slide 28

• relapse sbp age • anticipate yhat, xb • foresee rstudent, rstudent • diagram rstudent yhat, yline(- 3, - 2, 0, 2, 3)

Slide 29

Graphical Analysis of Residuals Studentized lingering (or folding blade leftover) versus can be utilized to recognize nonlinearity . Studentized leftover (or pocketknife lingering) versus (then again x ) can be utilized to distinguish heteroscedasticity . In STATA ® , charge "hettest" performs Cook-Weisberg test for heteroscedasticity after "relapse."

Slide 30

. relapse SBP Age Source | SS df MS Number of obs = 70 - + - - F( 1, 68) = 77.92 Model | 15068.9324 1 15068.9324 Prob > F = 0.0000 Residual | 13150.4391 68 193.38881 R-squared = 0.5340 - + - - Adj R-squared = 0.5271 Total | 28219.3714 69 408.976398 Root MSE = 13.906 - - - - - SBP | Coef. Sexually transmitted disease. Fail. t P>|t| [95% CI] - + - - - - Age | .9871668 .1118317 8.83 0.000 .764 1.210 _cons | 104.1781 5.422842 19.21 0.000 93.357 115.000 - - - - - . hettest Cook-Weisberg test for heteroskedasticity utilizing fitted SBP Ho: Constant change chi2(1) = 0.07 Prob > chi2 = 0.7975

Slide 31

Analysis of Leverage Points Observations compare to huge corner to corner components of the cap network (i.e., h ii >2(p+1)/n) are considered as use focuses.

Slide 32

. relapse sbp age Source | SS df MS Number of obs = 32 - + - - F( 1, 30) = 45.18 Model | 3861.63037 1 3861.63037 Prob > F = 0.0000 Residual | 2564.33838 30 85.4779458 R-squared = 0.6009 - + - - Adj R-squared = 0.5876 Total | 6425.96875 31 207.289315 Root MSE = 9.2454 -