marginallet美人3

Titlestata.
commkspline—LinearandrestrictedcubicsplineconstructionDescriptionQuickstartMenuSyntaxOptionsRemarksandexamplesStoredresultsMethodsandformulasAcknowledgmentReferencesAlsoseeDescriptionmksplinecreatesvariablescontainingalinearsplineorarestrictedcubicsplineofanexistingvariable.
Forlinearsplines,knotscanbeuserspecied,equallyspacedovertherangeofthevariable,orplacedatpercentiles.
Forrestrictedcubicsplines,alsoknownasnaturalsplines,knotlocationsarebasedonHarrell's(2001)recommendedpercentilesoruser-speciedpoints.
QuickstartLinearsplineofv1withknotsat30,40,and50mksplineknot130knot240knot350knot4=v1Addknotsat20and60mksplineknot120knot230knot340knot450knot560knot6=v1Deneknotsbyquintilesmksplineknot5=v1,pctileAsabove,butapplyfrequencyweightwvarbeforecalculatingquintilesmksplineknot5=v1[fweight=wvar],pctileRestrictedcubicsplineofv2withdefault5knotsmksplineknot=v2,cubicAsabove,butplaceknotsat30,40,and50mksplineknot=v2,cubicknots(30,40,50)MenuData>Createorchangedata>Othervariable-creationcommands>Linearandcubicsplineconstruction12mkspline—LinearandrestrictedcubicsplineconstructionSyntaxLinearsplinewithknotsatspeciedpointsmksplinenewvar1#1[newvar2#2newvark=oldvarifin,marginaldisplayknotsLinearsplinewithknotsequallyspacedoratpercentilesofdatamksplinestubname#=oldvarifinweight,marginalpctiledisplayknotsRestrictedcubicsplinemksplinestubname=oldvarifinweight,cubicnknots(#)knots(numlist)displayknotsfweightsareallowedwiththesecondandthirdsyntax;see[U]11.
1.
6weight.
OptionsOptionsmarginalisallowedwiththerstorsecondsyntax.
Itspeciesthatthenewvariablesbeconstructedsothat,whenusedinestimation,thecoefcientsrepresentthechangeintheslopefromtheprecedinginterval.
Thedefaultistoconstructthevariablessothat,whenusedinestimation,thecoefcientsmeasuretheslopesfortheinterval.
displayknotsdisplaysthevaluesoftheknotsthatwereusedincreatingthelinearorrestrictedcubicspline.
pctileisallowedonlywiththesecondsyntax.
Itspeciesthattheknotsbeplacedatpercentilesofthedataratherthanbeingequallyspacedovertherange.
nknots(#)isallowedonlywiththethirdsyntax.
Itspeciesthenumberofknotsthataretobeusedforarestrictedcubicspline.
Thisnumbermustbebetween3and7unlesstheknotlocationsarespeciedusingknots().
Thedefaultnumberofknotsis5.
knots(numlist)isallowedonlywiththethirdsyntax.
Itspeciestheexactlocationoftheknotstobeusedforarestrictedcubicspline.
Thevaluesoftheseknotsmustbegiveninincreasingorder.
Whenthisoptionisomitted,thedefaultknotvaluesarebasedonHarrell'srecommendedpercentileswiththeadditionalrestrictionthatthesmallestknotmaynotbelessthanthefth-smallestvalueofoldvarandthelargestknotmaynotbegreaterthanthefth-largestvalueofoldvar.
Ifbothnknots()andknots()aregiven,theymustspecifythesamenumberofknots.
Remarksandexamplesstata.
comRemarksarepresentedunderthefollowingheadings:LinearsplinesRestrictedcubicsplinesmkspline—Linearandrestrictedcubicsplineconstruction3LinearsplinesLinearsplinesallowestimatingtherelationshipbetweenyandxasapiecewiselinearfunction,whichisafunctioncomposedoflinearsegments—straightlines.
Onelinearsegmentrepresentsthefunctionforvaluesofxbelowx0,anotherlinearsegmenthandlesvaluesbetweenx0andx1,andsoon.
Thelinearsegmentsarearrangedsothattheyjoinatx0,x1,whicharecalledtheknots.
Anexampleofapiecewiselinearfunctionisshownbelow.
Example1Wewishtotamodeloflogincomeoneducationandagebyusingapiecewiselinearfunctionforage:lninc=b0+b1educ+f(age)+uTheknotsaretobeplacedat10-yearintervals:20,30,40,50,and60.
.
usehttps://www.
stata-press.
com/data/r16/mksp1.
mksplineage120age230age340age450age560age6=age,marginal.
regresslninceducage1-age6(outputomitted)Becausewespeciedthemarginaloption,wecouldtestwhethertheageeffectisthesameinthe30–40and40–50intervalsbyaskingwhethertheage4coefcientiszero.
Withthemarginaloption,coefcientsmeasurethechangeinslopefromtheprecedinggroup.
Specifyingmarginalchangesonlytheinterpretationofthecoefcients;thesamemodelistineithercase.
Withoutthemarginaloption,theinterpretationofthecoefcientswouldhavebeendydage=a1ifage00,ifu≤0Withouttheknots()option,thelocationsoftheknotsaredeterminedbythepercentilesrec-ommendedinHarrell(2001,23).
Thesepercentilesarebasedonthechosennumberofknotsasfollows:No.
ofknotsPercentiles3105090453565955527.
55072.
59565234159779572.
518.
3334.
175065.
8381.
6797.
5Harrellprovidesdefaultpercentileswhenthenumberofknotsisbetween3and7.
Whenusinganumberofknotsoutsidethisrange,thelocationoftheknotsmustbespeciedinknots().
AcknowledgmentTherestrictedcubicsplineportionofmksplineisbasedonthercsplinecommandbyWilliamDupontoftheDepartmentofBiostatisticsatVanderbiltUniversity.
ReferencesHarrell,F.
E.
,Jr.
2001.
RegressionModelingStrategies:WithApplicationstoLinearModels,LogisticRegression,andSurvivalAnalysis.
NewYork:Springer.
Newson,R.
B.
2012.
Sensibleparametersforunivariateandmultivariatesplines.
StataJournal12:479–504.
Orsini,N.
,andS.
Greenland.
2011.
Aproceduretotabulateandplotresultsafterexiblemodelingofaquantitativecovariate.
StataJournal11:1–29.
Alsosee[R]fp—Fractionalpolynomialregression