latinamerican macroevolution workshop unam 2018phytools.org/mexico2018/lec/pgls.pdf ·...
TRANSCRIPT
LatinamericanMacroevolutionworkshopUNAM2018
AlejandroGonzalez-VoyerInstitutodeEcología
UNAM
Altitude
SVL
s1s2s3s4s5s6s7s8s9s10s11s12s13s14s15s16s17s18s19s20s21s22s23s24s25s26s27s28s29s30s31s32s33s34s35s36s37s38s39s40s41s42s43s44s45s46s47s48s49s50
0
12.5
25
37.5
50
TypeIErrorSinfilogenia FilogenéJco
500simulationsofphenotypicevolutionfor2independenttraitsalongthephylogeny
%Significantcorrelations
Risksofignoringphylogeneticnon-independence
PhylogeneticGeneralizedLeastSquares(PGLS)
• Generalizedleastsquaresmodelwherethephylogeneticrelationshipsamongspeciesareincorporatedintheerrorterm
• Variancecovariancematrixdescribestheexpectedsimilarityamongspeciesbasedonthedegreeofsharedancestry
• Moreprecisely:expectedcovarianceamongresidualsduetosharedancestry(phylogeneticrelationships)amongspecies
Grafen1989;Martins&Hansen1997
Covarianceandphylogeneticrelationships
A
B
CD
EFG
H
I
J
L
K
A&B:Expectedcovariance=high
A&D:Expectedcovariance=low
A&K:Expectedcovariance=null
A B C D E F G H I J K LA δ2 + + + + - - - 0 0 0 0B δ2 + + + - - - 0 0 0 0C δ2 + + - - - 0 0 0 0D δ2 + - - - 0 0 0 0E δ2 - - - 0 0 0 0F δ2 + + 0 0 0 0G δ2 + 0 0 0 0H δ2 0 0 0 0I δ2 + + +J δ2 + +K δ2 +L δ2
Variancecovariancematrix
�8
A B C D E F G H A B C D E F G H
A B
CD
EF
GH
A
B
C
D
E
F G
H
• AdvantagesofPGLS:–Modeloftraitevolutionisdeterminedbythevariancecovariancematrix,hencecanaccomodatedifferentevolutionarymodels–Canincorporateanestimateofintraspecificmeasurementerrorinthemodel–Cancombinecontinuousanddiscretetraitssimplyinasinglemodel–Canobtaintheinterceptvalue
Phylogeneticgeneralizedleastsquares
• UseofPGLSdoesnot“remove”partofthecorrelationduetothephylogeny
• Theeffectoftakingphylogenyintoaccountistoreducethevariancearoundtheestimatedslope
• Theirmeansarenotbecausesinceestimatesofregressioncoefficientsareunbiasedwhetherornotthecorrectphylogenyistakenintoaccount.
Rohlf2006
ModelsofEvolution
• PGLSmethodestimatesanevolutionaryparametersimultaneouslywithmodelfit
• Evolutionaryparameter(λ,α)describesthemodelofevolution&adjuststhevariancecovariancematrixtothemodel– λ:adjuststhevariancecovariancematrixtoaBrownianevolutionmodel– 1:evolutionfollowsaBrownianmodel– 0:evolutiondoesnotfollowaBrownianmodel,thereisnophylogeneticsignalintheresiduals
PICvsPGLS• IfcovarianceintheresidualsfollowsastrictBrownianmodelPIC&PGLSshouldconvergeinthesameresult
• RegressionsthroughtheoriginofPICsareidenticaltothoseobtainedusingPGLSwithaninterceptincluded
• HoweverwhenevolutionisnotBrownianPGLSoutperformsPICs
• Maximumlikelihoodestimateoflambdaallowsfornecessarycorrectionforestimatedcovarianceofresiduals
Rohlf2001;Revell2010
The7deadlysinsofcomparativeanalyses
1. PuttingunduefaithinmodelswithlowR2–OnlyasinglepackageinRestimatesR2forphylogenticGLSmodels–SomequestiontheaccuracyofR2estimatesinphylogeneticGLS
Freckleton2009
2. Reportingtheresultsofphylogeneticallyindependentandphylogeneticanalyses:– thetwomodelsmakedifferentassumptions
aboutthedistributionofdata– atbestregardedasalternativemodelsofthe
samedata– assuchtheyshouldNOTbetreatedequally
3. Nottestingdistributionalassumptions:– Phylogeneticanalysesmakeassumptionsabout
thedistributionofresidualsthatarethesameasthosemadeinnonphylogeneticanalyses
– Roffersdifferentwaysoftestingresidualdistributions(incapersimplyuseplot()function)
Freckleton2009
4. Datadredging:– becausedataarenotgeneratedexperimentally,
therelationshipsfoundareonlycorrelative– possibilityofhiddenvariablesmeanthat
significantcorrelationsmaybemistakenlytakenasimplyingacausalrelationshipwhennosuchrelationshipexists
– shouldavoidthetemptationofincludingalargenumberofpredictorsto“seewhatcomesout”
Freckleton2009
5. Treatingresidualsasdata:– Useofresidualstocontrolforconfoundingvariablesis
wrong– Canresultinbiases– UseANCOVAtocontrolforallometriceffectsor
confoundingvariables6. Ignoringalternativemodels:– UsemodelcomparisonwithAICorAICC
7. Ignoringqualitycontrolofdata:– Datafromdisparatesourcesmaybeofdifferent
quality– Lowqualitydatacompromisesstatisticalpower– Missingdatacanleadtobiasesintheoutcomeof
analysesFreckleton2009
Correlatesofspeciesrichness
• Intrinsicorextrinsicspeciescharacteristicsmayinfluencespeciationrates
• CanusePGLSmodelstoanalyzethecorrelationbetweenspeciescharacteristicsandspeciesrichness
• Dependentvariableinsuchmodelsisrateofdiversificationorspeciesrichness