Thanks to visit codestin.com
Credit goes to www.scribd.com

Scribd
Upload
17 views12 pages

MC Methods of Detection

method of detection of multicollinearity

Uploaded by

ps9366610862
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
17 views12 pages

MC Methods of Detection

method of detection of multicollinearity

Uploaded by

ps9366610862
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 12
Econometric Problems: 5 238 Seong Fey » r, in most cases the whole soy Ct Of @ icula ‘ rch, In partic ometric researe Ag it has become a common practice , Play 0 “prog, Maloy econ ee not variables is RO - | ae periment ion, that 18, by gradually including additional y My Yahi and judging their importance by using, among othe; ‘*Tableg gt ese errors For example suppose that the true relationsht’ Ship jg their standar Yedo thi Xi + baX, tu wn, the researcher usually 1 starts h ons Since the true specification is not kno aaa he sintpler tentative formulation study with ¢ YeantaX, tv y yield significant results, although a, wif : sh ay will he he addition of X, i ‘i haw 8 he ince the model is in this ¢ case This model will most probably yi¢ ror due to the omission of X) cification en relationship should normally improve the fit correctly specified. However, if X1 and X, are highly correlated and th standard errors are large, the researcher will usually reject X2, being mi ‘ his case multicollinearity results in the a by its large standard error. Int fication of the model, since Xy is by decision and hence in the wrong speci assumption (that is in the postulated model) an important explanatory variable in the relationship. (See Theil, Economic Forecasts and Policy, p. 217, See also Farrar and Glauber, op. cit., P- 94.) However, large standard errors do not always appear even in functions in which the regressors are strongly multicollinear. For example production functions with overall correlations much in excess of 0-950 have been well- estimated with intercorrelations between labour and capital as high as 0-800 to 0-900. If these functions were not well estimated, we would tend to find high sampling errors of the estimates coefficients. By conventional criteria the estimated parameters of most Cobb—Douglas production functions ace large selative to their standard error. The coefficients are generally high multiples of a errors. (See L. R. Klein, Introduction to Econometrics, P. tly S. D. Silvey has published a study on the problem of multicollineatit! Bae Silvey, “Multicollinearity and Imprecise Estimation’, Roya! Statist ty, 1969.) We will not examine Silvey’s approach here, since it Is not oe tially superior to Farrar's and Glauber’s test, which will be develOP® 2 next section. 11.4, TESTS FOR DETECTING MULTICOLLINEARITY moarit? 239 dard errors do ‘ , bb Douglas oo appear with multic ast arise for beis rears em: Pitesti (see S 4 large sta at tionships among the explanatory eae because of the Neen The intercorrelations of the explanator ables, . a ithe b's and theit standard ci variables need not be high { tan adequate criterion by itself ne . , is a ec vi ty 1 The © erall R? may b high (relative to the r. 0 ie : ‘x; x/'8) and yet the results may pohighY imprecise and insignificant (with wrong signs and a. ee . and/or large standard weer. a combination of all these criteria me aioli. In order to gain as much Read help the detection of -osnesS of multicollinearity we suggest the « sae as possible as to the Seep essence See ieediversion of a eae of the following method "Map Analysis’). sch’s ‘Confluence Analysis’ (or The procedure is to regress the depe aria tory variables separately. Thus aan a ana and we examine their results on the basis of a ae cena pees We choose the elementary Perion whic a statistical criteria. results, On both a priori and statistical criteria, ea . Ea Sno variables and we examine their effects on the ere ae Se aoe sandard errors, and on the overall R?. A new variabl ‘i ot ae ae fous or detrimental, as follows: SEE a : ue, new variable improves R? without rendering the individual ee ona priori considerations, the variable is considered ained as an explanatory variable. (2) If the new variable does not improve R? and does not affect to any =. extent the values of the individual coefficients, it is considered as 8) a ae ae ee as the coer variables). Peis es ynsi rably the signs or the values of the . nts, it is considere as detrimental. If the individual coefficients are f affected in such a way as to become unacceptable on theoretical, ¢ priori, cman then we may say that this is warning that multicollinearity is ; z problem. The new variable is important, but because of intercorrela- : with the other explanatory variables its influence cannot be assessed i tically by ordinary least-squares. This does not mean that we must reject detrimental variable. If we did so, we would ignore information valuable to es of approaching as best we can the

You might also like