Page 30 - Regression Guideline
P. 30
Important"Terms"You"Need"to"Know"
EsDmated&Sales&price&=&30,000&+&100(sq&foot)&+&4,000(lakefront&property)&&
"
The"coefficient"for"square"foot"is"100"and"the"coefficient"is"4,000"for"lakefront"property.""This" means"the"predicted"sales"price"increases"$100.00"for"every"1Pfoot"increase"in"square"footage" (a"conJnuous"variable"with"a"range"of"values)"and"an"addiJonal"$4,000"if"the"property"is"also" on"a"lakefront"(a"discrete"variable"that"can"either"be"present"or"absent).""Using"this"equaJon," the"esJmated"sale"price"for"a"1,000"square"foot"home"on"a"lakefront"would"be"$134,000." (30,000"+"100,000"+"4,000)."If"the"home"were"not"on"the"lakefront,"its"appraised"value"would" drop"to"$130,000."""
"
Do"not"confuse"the"size"of"the"regression"coefficient"for"its"relaJve"importance"in"the"model." This"is"determined"by"the"amount"of"variaJon"in"sale"price"the"property"characterisJc"accounts" for"(RPsquare,"see"below)"and"its"staJsJcal"significance"(FPtest)."SomeJmes,"counterPintuiJvely," large"coPefficients"are"not"meaningful"while"small"ones"are."Regression"coefficients"can"be" either"posiJve"or"negaJve."PosiJve"coefficients"are"associated"with"increases"in"sales"price"as" the"value"or"their"corresponding"property"characterisJc"increases."NegaJve"coefficients"mean" the"value"of"a"property"decreases"as"their"value"increases"or"by"being"present"(e.g.,"close" proximity"to"a"noisy"boulevard).""
&
R4Squared&(R2):&is"a"very"helpful"and"intuiJve"measure"of"how"much"your"regression"model" accounts"for"variaJon"in"the"sales"prices"among"the"properJes"analyzed."Values"can"range" from"0"to"1.00"(but"rarely"reach"these"limits)"with"larger"values"indicaJng"a"beser"model"fit"or" a"higher"amount"of"the"price"variaJon"accounted"for"by"the"predictors"in"the"model."An"R2"of". 80,"for"instance,"means"the"set"of"property"characterisJcs"included"as"predictors"in"the" regression"equaJon"accounts"for"80%"of"the"variaJon"in"known"sales"prices"among"the" properJes"examined"in"the"analysis."That's"an"excellent"result"because"it"means"if"you"know" the"corresponding"values"for"the"characterisJcs"of"the"property"you"are"appraising,"you"can" likely"appraise"the"value"of"that"property"with"good"accuracy"based"on"how"well"the"model" predicted"the"values"of"other"properJes"recently"sold."
""
29"