Now that you've seen the results of the sprinkler data analysis, what do you recommend that marketing management do with the less than spectacular results? What other analyses would you recommend? What data do you wish you had?

Words required:5-1000Words written: 0

• visibility
• Answers are visible to other students
• lock
• You cannot edit your answer(s)

Submit Answer

.

Let'scontinueourwalkthroughtheworldoflogisticregressionwithanewdatasetandalittlebitdifferentkindofanalysisthatwe'regoingtobedoing.Iwanttoshowyouhowsometimesusingdifferentdatatypescanproducevery,vastlydifferentresultsusinglogisticregressionandsometimesresultsthatarekindofdifficulttointerpret,atleastinhardquantifiableterms.Thisisaninterestingdatasetthatalsoshowsthatsometimestheoverallmodelingprocessdoesn'treallytakeyouanywhereveryspecific.

Let'stakealookatthedatabydescribingitfirstandwhatitrepresents.Thisisdatafromanationalchainofcompaniesthatselllawnirrigationequipment.Andwhattheywantedtodowasbeabletopredictthefactorsthatmostpredictwhetherornotafacilitiesmanagerorothergroundskeeperwouldbeinterestedinbuyinganirrigationsystemorasprinklersystemfortheirproperty.

Now,thisisn'tforresidentialit'sforcommercial.Andsotheyconductedasurveyofotherfacilitiesmanagersofdifferentkindsofcommercialfacilitiesandgatheredsomeinformationfromthemandtriedtouseitinalogisticregressiontoexplainwhetherornottherewasactuallyasprinklersysteminstalledonthepropertygroundsorwhethertherewasn't.Thehopewastobeabletoidentifythefactorsthathadledpeopleinthepasttoinstallsprinklersystems,andthentobeabletodevelopacustomerprofileofthosewhowouldbemostlikelytobuybasedonthepropertytypeandsomeothervariables.

Sothroughthissurvey,theyaskedseveralquestionsthatIhavesortofidentifiedhereinthisdata.Firstisthefacilitytype.Thatis,whatwasthemainuseofthefacilitythatthefacilitiesmanagerworkedat?Wasitprimarilyofficeorretail?Orwasitprimarilyaprimarilywarehouseorindustrial?Noticethatthisisacategoricalvariable.It'skindoflikesex,numberreallydoesn'tmeananythinghere.Sowe'llhavetotreatitassuchwhenwegetreadytomodelit.

Nextwasascaleditemofthefacilitiesmanagerabouthowimportantitwastothemorthebuilding'sownerabouttheappearanceofthebuildingtothegeneralpublic.Becausesomebuildings,particularlyifthey'reheavyindustrialorsomethinglikethat,appearanceandlandscapingisnotsomethingthat'sterriblyimportanttoallbuildingtypes.Whereasifyouhavearetailbuilding,forexample,thenitmightbe.Sowhetheritwasnotatallimportantuptoveryimportant,soordinaldatatobeusedhere.

Nextwasthegreenspacesize.Now,greenspacesizediffersagreatdeal.Anditdiffersalongacoupleofdimensions.What'sthepercentageofthetotallotsizethatisgreenspace?Andwhat'sthetotalsizeofthelot?Someasuringthisusingratiodatawouldhavebeenkindofdifficult.Icouldhavealotofgreenspaceonasmalllotthat'salmostallspace.

Someasuringitinsquarefeetoranythinglikethatrelativetolotsizewouldhavebeenkindofdifficult.Sooneofthethingsthatwedidinthisparticularsurveywastoasktheminanordinalnaturewhetherornottherewaslittle or nogreenspace,justaslightamount,amoderateamount,averylargeamount,andsoforth.Sowetookthisthingthatnormallywouldbeexpressedinsomethinglikesquarefeetoracresandwehadthemtranslateitintowhethertheyfeltlikeitwasalittleoralot.

Nextwasthesquarefootageofthebuildingsthattheymanagedatthatfacilityonascalefrom1to6dependingonbuildingsize.I'vecodeditasordinaldata,itwon'tmakeanydifference.Youcouldreasonablysaythatit'sratiodata,butitwon'tmakeanydifferencetotheanalysis.Butitwillmakealittledifferencetotheinterpretation,soI'llbesureandpointthatout.

Nextisthenumberofvisitorstothepropertyperdaywith1beingvery fewerthan100,morethanbeing5,000ormoreinthecaseoflargeretailoperations,andthenfinallyaddedsomeadditionaldatabasedonthezipcodefromwhich thefacilitywaslocatedastowhattheaverageannualrainfallinthatparticularareawas,whichobviouslymightalsoaffectwhetherornotwewantalawnirrigationorsprinklersystemonourcommercialproperty.

Sokindofalongwindedexplanation,butIthinkit'simportanttogetanideaofthekindofdatathatyou'reusing.We'regoingtorunthroughtheoutputofthelogisticregressionprettyquickly,becausewespentalotoftimeonitwhenwewerelookingatthecellphonedata.AndIjustwanttokindofgetdowntothenittygrittyofthisandshowyousomeimportantpointsaboutit.

Sowe'regoingtogotoRegression,BinaryLogistic.We'regoingtoaskastheSPSStodesignatethis--isthesprinklersysteminstalledonthepropertyornot--asourdependentvariable.Andwe'regoingtousealloftheseothervariablesasourindependentvariables,notourcovariants,butareindependentvariables.Andthenwe're going todesignatethemainuseofthefacilityasacategoricalvariable.AndI'llleaveitaslast.Ifitbecomesdifficulttointerpret,ifIneedto,thenIcanchangeittofirstlaterintermsofidentifyingmyreferenceandcomparisonlevelofcategory.

Allright,that'sallIneedtodo.Let'sgoaheadandclickOK.Andwe'renotgoingtotalkaboutallofthesedifferentdiagnosticinformationatthefront.We'resimplygoingtogettoblockone,whichistheonewe'reinterestedin,andwe'regoingtotakealookatsomeofthedifferentmetricsofsuccesshere.OurNaglekerkeRsquareis0.20,notasgoodaslasttime,butit'sallright.

Itdiddoafairlygoodjobofclassifyingresultshereuptoaround70%,soI'mnotunhappywiththat.Anditwasareasonablygoodintermsofclassifyingwhetherornotithadnosysteminvolvedversushavingasysteminvolved,butthemodelwasfarbetteratpredictingwhethertherewasasystemrelativetowhethertherewasn'tasystem.Andthatlittlebitof asymmetryisnotabigproblem,butwesurelikeitwhentheycancategorizewellbothsidesofthebinarydependentvariable.

Anddownhere,wehavewheretheactionis,whatwewanttolookatintermsofthemodelingprocess,andjustbeabletopickoutthosevariablesthatarestatisticallysignificant,thosethatarenot,andbeabletoeliminatefromthemodelthosethatarenotsowewindupwithaniceparsimoniousfinalmodel.Well,facilitytypegreaterthan0.5,sowe'llgetridofit.We'llgetridoftheappearanceimportance,becauseitwasn'tsignificanteither,whichsurprisedmealittlebit,butitdidn'tcrackoutsothere'snoreasontokeepit.

Wearegoingtokeepthe greenspacesizewhichseemstomatter.Andthatseemstomakesensetome.Thesizeofthebuildingdoesn'tmatter.Andthatactuallymakessense,becausealotoffacilitieshavealargegroundswithhugelongspaceinonlyasmallbuilding.Otherlocationsmayhavethebuildingtakingupalmosttheentirelot.Sothebuildingsizemaynotalwaysbereallydependentorrelatedtotheoverallsizeofthegreenspace.Sothisisnotonethatsurprisesmealtogether.Thenumberofvisitors?Thatoneactuallydoesn'tsurprisememucheither.Itwasjustavariableweincludedsimplybecauseitseemedtomakesenseandmighthaveagoodpossibilityofincludingitandcollectingtheinformationwasn'treallyallthatdifficulteither.

Annualrainfall?Obviously,that'sanimportantvariableanditcamebackashighlysignificant.Sowereallyonlyhavetwothatwewanttofocusonatthispoint--theamountofrainfallandthe greenspacesize.Therestofthesesortoffalloff,allright?Solet'sgoaheadandtakealookatthesedatabygoingbackandlookingatthe greenspacesizeandtheannualrainfallandseeingwhatourmodellookslikeattheend.

Sowe'llgothroughandwe'llhitourbinarylogistic.Andwe'lleliminatealloftheseothervariablesthatwerenon-significantfromthemodel.Facilitytypegoes.Theappearanceimportancegoes,whichreallykindofsurprisedme.Thebuildingsizegoes.Thenumberofvisitorsgoes.Andwe'releftwithtwovariableshere.Let'sgoaheadandclickOK.

Andhere'stheresultsthatwegetaswescrolldown.Noticethatitstillcorrectlycategorizesabout70%ofthem,buttheproportionofcorrectcategorizationsisaboutthesameaswell.NoticethatourNaglekerkeRsquareddidnotchange,Ithink,atall.Ifitdid,itwastinyandnoticethatallofourvariablesnow,orbothofthem,arestatisticallysignificant.

Solet'stakealookatinterpretingsomeoftheseintermsofouroddsratiosandseewhattheymean,allright?Now,inparticular,I'minterestedininterpretingthe greenspacesize.Whatdoesthatmean?Doesitmeanthatforeveryoneunitincreaseinspacesize,thentheoddsofneedingorhavingasprinklersystemgrowbyabout1.145times?

Allright,fine,butwhatdoesthatmean?Becauseifyou'llgobackandyou'lllookatthedata,thesizeofthegreenspace--aoneunitincrease,whatdoesthatmean?AsImovefromlittleornogreenspacetoaslightamounttoamoderateamount,usingthiskindofperceptualdataisusefulfortellingusthatgreenspaceisimportanttopredictinghowmuchofwhetherornotanareahasasprinklersystemornot,butbytheamountandactuallymakingsenseoftheoddsratiowhenyouusethosekindsofordinaldata,notsomuch.

Soeventhoughitisstatisticallysignificantasindicatedbythep-value,becausewe'veusedordinaldata,thisdoesn'treallymakeawholelotofsense.Theannualrainfalldoesmakemoresense,becauseit'sratiodataandit'sshowingusthat,foreveryinchofchangeinrainfall,theoddsgobyabout0.55to1.Andnoticethatit'snegative,andthatmakessenseaswell.Thatis,themorerainfall,thelessIneedanewsprinklersystemorwanttoinstallasprinklersystem,becauseIgetplentyofrain.ThelessrainfallIhave,themoreIneedasprinklersystem.ThemorerainfallIhave,thelessIdo.Itmakesperfectsense.

Sonow,let'skindoflookandseewhatwe'vedone.Westartedoffwithalotofdifferentvariables,mostofwhichwewoundupeliminating.Andthetwovariablesthatwehaveleftmakesensetohave--theamountofgreenspace,evenperceptuallytheamountofgreenspaceontheproperty,andtheamountofrainfallthatthepropertyhasonit.

Doesthatreallytellusanythingthatwedidn'tknowtostartwith?Doesthisreallygiveusagreatinsightintowhyitisthatpeoplewouldhaveasprinklersystemornot?Doesitreallyhelpusachievethegoalofthedata?Andthatispredictingthecustomertypeswhoaregoingtobemostlikelytowantsprinklersystemsontheircommercialpropertyinthefuture.Theanswerisreallynotawholelot.Itwouldmakesensetomethatifacompanyhadalotofgreenspaceinanareathatdidn'tgetalotofrain,theywouldbeinthemarketforasprinklersystemiftheydidn'talreadyhaveone.

Sosometimes,thesecomplicatedanalyzeswindupleadingyoutoconclusionswhereyousortofshrugyourshouldersandsay,well,youknowwhat?Itwasanicetry.AndmaybethereareotherthingsIcandowiththedata.Andinfact,inthiscase,therewere.Butusingalogisticregressionmodelwiththevariablesthatwehadatourdisposaltopredictwhoisorwhoisnotgoingtoneedagreenspacebasedontheinformationthatwehad--itdidn'tturnouttoproduceawholelotofinsightsmorethanweexpecteditto.

Butwhenthosekindsofanalyzesdo,t

heycanbereally,reallyhelpful.ButIdowanttosetyourexpectationsbysayingthatsometimesevenwell-conductedanalyzeswillleadyoutoaplacewhereyousay,well,youknowwhat?Ikindofalreadyknewthat.That'sOK.Marketinganalyticsisn't100%rightallofthetime.Andwhatyoudoisyousay,youknowwhat?Onbalance,it'smorevaluabletodoitthanitistonot.