【正文】
ictor Coef SE Coef T P Constant Score1 選擇統(tǒng)計(jì)回歸回歸 ,輸入Score2,在預(yù)測(cè)因子欄中,輸入Score1. “選項(xiàng)”按鈕。 2. 1.您可以檢驗(yàn)兩組數(shù)據(jù)之間的關(guān)系看看是否可由容易獲得數(shù)據(jù)來預(yù)測(cè)想要的數(shù)據(jù)。 Cook’s distance bines leverages and Studentized residuals into one overall measure of how unusual the predictor values and response are for each observation. Large values signify unusual observations. Geometrically, Cook’s distance is a measure of the distance between coefficients calculated with and without the ith observation. Cook [7] and Weisberg [29] suggest checking observations with Cook’s distance F (.50, p, np), where F is a value from an Fdistribution. n DFITS, like Cook’s distance, bines the leverage and the Studentized residual into one overall measure of how unusual an observation is. DFITS (also called DFFITS) is the difference between the fitted values calculated with and without the ith observation, and scaled by stdev ( i). Belseley, Kuh, and Welsch [3] suggest that observations with DFITS 2 should be considered as unusual. See Help for more details on these measures. Example of performing a simple linear regression 簡(jiǎn)單線性回歸實(shí)例 您是一個(gè)制造者并想要容易地獲得一個(gè)產(chǎn)品的質(zhì)量標(biāo)準(zhǔn),但是該程序十分昂貴。MINITAB將這些值在高leverage異常觀測(cè)值表中顯示。,其中X是設(shè)計(jì)矩陣,其中hi僅與預(yù)測(cè)因子有關(guān),它與響應(yīng)Y有關(guān)。這三種方法是:Leverages、Cook’s distance,及DFITS Leverages是“hat”矩陣的對(duì)角,H = X (X162。如果您使用帶權(quán)重的預(yù)測(cè),可以參考幫助中的獲得正確的結(jié)果。如果輸入了常數(shù)和一列,MINITAB會(huì)認(rèn)為您想要得到常數(shù)和每列數(shù)據(jù)組合的所有預(yù)測(cè)值。輸入常數(shù)或包含新X值的列,每個(gè)預(yù)測(cè)因子數(shù)據(jù)應(yīng)是一列(one for each predictor)。For each predictor, a curvature test and an interaction test are performed by paring the fit above and below the predictor mean using indicator variables(對(duì)于每個(gè)預(yù)測(cè)因子,可以用曲率檢驗(yàn)和交互檢驗(yàn)檢驗(yàn)通過使用指示變量業(yè)比較擬合度是高于并低于預(yù)測(cè)因子平均值) 也可以用另一個(gè)試驗(yàn)通過將關(guān)系模與數(shù)據(jù)“中心”部分?jǐn)M合,然后比較中心數(shù)據(jù)誤差平方和所有數(shù)據(jù)誤差平方和。參考[6] 和“幫助”得到更多的信息。如果需要其它信息,請(qǐng)參考[9], [22], [29]. 數(shù)據(jù)子集lackoffit檢驗(yàn) MINITAB同樣也可以進(jìn)行l(wèi)ackoffit檢驗(yàn)數(shù)據(jù),其數(shù)據(jù)不需要副本但是要包含數(shù)據(jù)子集。誤差項(xiàng)將被分成純誤差(error within replicates)和lackoffit誤差。 the data subsetting lackoffit test does not require replicates. 如果想得到其它信息,請(qǐng)參考[4], [22]. 檢驗(yàn)lackoffit MINITAB提供了兩種lackoffit 檢驗(yàn),這樣您可確定建立的回歸模型是否能夠完全適合您的數(shù)據(jù)。如果列中有丟失的觀測(cè)值,同樣在計(jì)算時(shí)這些數(shù)據(jù)就會(huì)忽略了,僅僅使用沒有丟失的數(shù)據(jù)。 。選定一個(gè)模型后檢驗(yàn)關(guān)系模型的假設(shè)是回歸分析的一個(gè)很重要的部分。 如果相互獨(dú)立的假設(shè)被破壞,一些關(guān)系模型的擬合結(jié)果就會(huì)被懷疑。 你應(yīng)該考慮使用選項(xiàng)中的方法來分散預(yù)測(cè)因子間的多重共線性: 重新搜集數(shù)據(jù),刪除預(yù)測(cè)因子,使用不同的預(yù)測(cè)因子或最小二乘法回歸分析的替代,獲得附加的信息請(qǐng)參考[3], [21].。VIF=1時(shí)表明因子之間不相關(guān),所有預(yù)測(cè)因子中最大的VIF通常是用來作為多重共線性的指示。 變量inflation factor The variance inflation factor (VIF) 用來檢測(cè)一個(gè)預(yù)測(cè)因子和剩下的預(yù)測(cè)因子是否有很強(qiáng)