测量的基本概念及理论.pptx

大綱大綱量化構念古典測量理論(classical measurement theory)共同因子(common factor)的概念 效度(validity)的相關概念-構念關係網(nomological network)-多元特質和多重方法矩陣(multitrait-multimethod matrix)測量尺度測量尺度(Measurement Scale)數字的精確程度:(1)類別尺度(Nominal Scale)(2)等級尺度(Ordinal Scale)(3)等距尺度(Interval Scale)(4)等比尺度(Ratio Scale)量化離職意向的構念量化離職意向的構念 請您圈選您對以下描述的同意程度：1=極不同意；2=不同意；3=沒所謂同意或不同意；4=同意；5=極同意-我常想到辭職。-我很可能于明年另尋新的工作。-如果能自由選擇，我不會喜歡留在這機構工作。古典測量理論古典測量理論(classical measurement theory)Observed Score(OS)受三個影響:(1)真實得分(True Score;TS)(2)獨特得分(Unique Score;US)(3)誤差得分(Error Score;ES)離職意向的例子:OS1=TS+US1+ES1(第一題)OS2=TS+US2+ES2(第二題)OS3=TS+US3+ES3(第三題)變異量變異量(Variance)及及共變量共變量(Covariance)-以變異量及共變量驗證構念間之關係-樣本整體變異量(Observed Variances;O),包括:(1)真實差異(True Variance；T)(2)獨有因素帶來的差異(Unique Variance；U)(3)隨機誤差帶來的差異(Error Variance；E)O=T+U+E信度信度(reliability)(1)E 佔 O 的比重(2)因為 E 是隨機的，信度為測量的工具免於 隨機誤差的程度(3)測量結果的一致性或穩定性 信度係數的估計信度係數的估計:為了與統計上的相關係數看齊，我們一般會取兩次測量的共變量比例的平方根，來合計信度，稱之為信度係數(reliability coefficient)。E1E2T+U信度係數的方程式信度係數的方程式信度係數=信度信度(reliability)的估計的估計-兩次測量的相關係數:（a）再測信度(test-retest reliability)（b）複本信度(alternative forms reliability)（c）折半信度(split-half reliability)（d）項目間的一致性(internal consistency reliability):Coefficient alpha;(SPSS“reliability”的指令)-一般來說信度係數要在0.7以上信度對檢定構念關係的影響信度對檢定構念關係的影響兩個構念的共變量(假設最理想情況):兩構念在測量時的獨有變異量均為零T1p+E1T2p+E2C兩個構念觀察所得的相關係數兩個構念觀察所得的相關係數兩個構念真實的相關係數兩個構念真實的相關係數測量工具的信度係數測量工具的信度係數 把把R0除以除以r1及及r2的平方根，的平方根，便可求得便可求得Rt Rt 與與 Ro的關係的關係(Correction for Attenuation)影響信度的主要因素影響信度的主要因素(1)受測量者方面(2)主持測量者方面(3)測量內容方面(4)測量情境方面(5)時間影響方面 共同因子共同因子(common factor)的概念的概念Common Factor=CFEF1+UF1EF2+UF2EF3+UF3OS1OS2OS3CF共同因子共同因子(common factor):共變量共變量T=共同因子的變異量(各項目的共變量:沒有測量誤差的變異量)C13C12U1+E1C23U2+E2TU3+E3因子的設定因子的設定:項目的加權總和項目的加權總和(Linear Combination)負荷量負荷量(L值值;factor loadings)及及固有值固有值(eigenvalues)(1)每一因子抽取了這九個項目總體變異量的 一部 分(2)每一因子抽取了九分一的總體變異量，由於不同情況牽涉的項目數不一樣，所以我們把這平均值標準化，以1為代表，稱為固有值(eigenvalue)(3)固有值為該因子的所有L值平方的總和(4)因子抽取的總變異量的比率便是固有值除以測量項目的總數(5)負荷量:由-1 到 1因子的數學背景因子的數學背景如有兩個項目(V1及V2)，變異量為VA1及VA2，共變量為CA，它們的加權總和(F)的變異量(FA)的方程式是：FA=(L1)2*VA1+(L2)2*VA2+(L1)2*(L2)2*CA(L1及L2小於1，所以FA是包含V1及V2部分的VA1、VA2及CA)。把共變量愈大的項目在同一個因子中的 L 值加大，這因子便能夠同時把這兩個項目更多的變異量和共變量都抽取。盡量用小數的因子來抽取最多的整體變異量，便會是相互間共變量大的項目在同一因子中的 L 值較大，而共變量大也符合了它們可能受同一原因影響的假定。因子分析的假設因子分析的假設(1)各因子抽取了總體變異量的不同部分(2)儘量用較少的因子來抽取最大比率的總體變異量;假設其他因子代表 U 及 E 的變異量(3)集中在少數因子，使少數項目的L值盡量擴大(rotation):(a)決定集中在多少個因子(b)是否容許這少數因子有相關探索性因子分析法的例子探索性因子分析法的例子(Exploratory Factor Analysis；EFA)182名香港的中學教師 三題測量同意特質(A1、A2及A3)、三題工作滿意度(JS1、JS2及JS3)、及三題自評的工作表現(JP1、JP2及JP3)EFA的的SPSS指令指令在沒有特定原因的情況下，我們以固定值等於1來抽取少數因子；另外，由於老師的同意特質性格可能與其工作滿意度及自評的工作表現有關，因此在rotation時我們不應假定因子之間是無關的，所以用SPSS的指令是：get file=name of file containing the SPSS save file.factor vars=A1 A2 A3 JS1 JS2 JS3 JP1 JP2 JP3/extraction=paf/rotation=oblimin.探索性因子分析法的結果探索性因子分析法的結果探索性因子分析法的限制探索性因子分析法的限制(1)L值的設定及最後抽出固有值大於1(或其他設定的標準)的因子數目，是完全取決於我們在實證研究所取得樣本資料(2)雖然我們說第一個因子的主要變異量是來自工作滿意度的三個測量項目，但它還是包括了其他六個測量項目的部分變異量，嚴格來說，這是不正確的 確認性因子分析法的假設關係確認性因子分析法的假設關係(Confirmatory Factor Analysis；CFA)C1C2C3確認性因子分析法的步驟確認性因子分析法的步驟F1=L11*JS1+L12*JS2+L13*JS3F2=L21*A1 +L22*A2 +L23*A3F3=L31*JP1+L32*JP2+L33*JP3-其餘的其餘的 L 值為零值為零-先對這些構念和它們的測量項目有一先對這些構念和它們的測量項目有一 個清楚及符合測量理論的關係假設，個清楚及符合測量理論的關係假設，然後以實證的樣本資料來驗證這關係然後以實證的樣本資料來驗證這關係假設假設 LISREL(CFA)指令指令Observed Variables:A1 A2 A3 JS1 JS2 JS3 JP1 JP2 JP3Latent Variables:AGREE JOBSAT JOBPERFRaw Data From File:(data file name)Relationships:A1 A2 A3=AGREEJS1 JS2 JS3=JOBSATJP1 JP2 JP3=JOBPERFEnd of ProblemCFA 結果結果因子的相關(Factor Correlation Matrix)AGREEJOBSAT JOBPERFAGREE 1.00JOBSAT 0.191.00JOBPERF 0.130.421.00檢定是否接受原來關係假設的標準檢定是否接受原來關係假設的標準指標指標(Goodness of Fit Statistics)(1)常用:RMSEA(或RMR;最好是少於0.08)、NNFI(也稱為TLI；最好是大於0.90)和CFI(最好是大於0.90)。(2)我們一般會報告Chi-Square及其Degrees of Freedom(它們的比率最好少於2.5)、及GFI(最好是大於0.90)(3)Chi-Square和GFI與樣本數有很大的關係，很多時樣本數愈大，它們反而更不理想，所以相對而言，RMSEA、TLI和CFI在判定是否接受原來關係假設更為重要。效度效度(validity)的概念的概念(1)測量結果是否正確，便是效度(validity)的問題(2)從O=T+U+E的方程式來理解，效度是指 T 佔 O 的比重(3)信度是效度的必要條件，而非充分條件(4)檢定效度的主要方法，最好是檢定構念間的關係網(nomological network)檢定效度的方法檢定效度的方法(1)內容效度content validity)(2)效標關聯效度(criterion-related validity):-同時效度(concurrent validity);-預測效度(predictive validity);-增加效度(incremental validity)(3)建構效度(construct validity):-輻合效度(convergent validity);-辨別效度(discriminant validity)多元特質和多重方法矩陣多元特質和多重方法矩陣(multitrait-multimethod matrix)多元特質和多重方法矩陣的分析多元特質和多重方法矩陣的分析(1)比較各相關係數的大小(2)CFA:C1的因子:M1C1,M2C1,M3C1C2的因子:M1C2,M2C2,M3C2C3的因子:M1C3,M2C3,M3C3M1的因子:M1C1,M1C2,M1C3M2的因子:M2C1,M2C2,M2C3M3的因子:M3C1,M3C2,M3C3影響效度的主要因素影響效度的主要因素(1)測量組成方面(2)測量實施方面(3)受測者反應方面(4)效標方面(5)樣本方面 建立測量尺度的例子建立測量尺度的例子-情緒智能情緒智能(1)從文獻及邏輯推論中確立構念的定義和範圍(Definition and Domain of the Construct)；(2)證明發展測量工具的必要性；(3)發展測量的項目(Item Generation)；(4)選擇測量項目(Item Selection)；(5)測量工具的效度檢定(Construct Validation)。EI 構念在文獻中的混亂構念在文獻中的混亂Different researchers have used different definitions of EISome researchers include not only emotion and intelligence,but also motivation,non-ability dispositions and traits,etc.For scientific construct,EI has to concentrate on“ability to handle or deal with emotions”EI 構念的定義和範圍構念的定義和範圍(1)Appraisal and expression of emotion in the self(Self Emotional Appraisal,SEA)This relates to the individuals ability to understand their deep emotions and be able to express these emotions naturally.People who have great ability in this area will sense and acknowledge their emotions well before most people.EI 構念的定義和範圍構念的定義和範圍(2)Appraisal and recognition of emotion in others(Others Emotional Appraisal,OEA)This relates to the individuals ability to perceive and understand the emotions of the people around them.People who are high in this ability will be much more sensitive to the feelings and emotions of others as well as reading their minds.EI 構念的定義和範圍構念的定義和範圍(3)Regulation of emotion in the self(Regulation of Emotion,ROE)This relates to the ability of a person to regulate their emotions,which will enable a more rapid recovery from psychological distress.A person with great ability in this area maintains positive emotions most of the time.As a result,this dimension is sometimes referred to as“self-motivation”or the ability of a person to be self-encouraging and be positive to life stresses.EI 構念的定義和範圍構念的定義和範圍(4)Use of emotion to facilitate performance(Use of Emotion,UOE)This relates to the ability of a person to make use of their emotions by directing them towards constructive activities and personal performance.A person who has high ability in this aspect is able to keep their behavior under control when they have extreme moods.The person will also make the very best use of their emotions to facilitate job performance in the workplace.證明發展測量工具證明發展測量工具(EI)的必要性的必要性(1)這構念在建立理論架構及增進我們知識領域的工作中是重要的:H1:Emotional intelligence is positively related to job performance.H2:Emotional intelligence is positively related to job satisfaction.H3:Emotional intelligence is positively related to organizational commitment.H4:Emotional intelligence is negatively related to turnover intention.證明發展測量工具證明發展測量工具(EI)的必要性的必要性(2)文獻中沒有現存適當的工具-Carson,Carson and Philips(1997)developed a 14-item measure of EI-Trait Meta-Mood Scale-Bar-On EQi instrument-Golemans 10-item measure-Multi-facet Emotional Intelligence Scale(MEIS;later version:MSCEIT)發展測量的項目發展測量的項目(Item Generation)定性(Qualitative)的 Asking managers and students to generate items to capture the construct(n=120)First introduced to the four dimensions of emotional intelligenceThey were then asked to generate self-reported items on each dimension that would describe a person with a high level of emotional intelligence.Preliminary deletion of(1)overlapping or similar items;(2)items with unclear meaning;and(3)items that did not match the definition of emotional intelligence Resulted in a 36-item preliminary measure of EI.選擇測量項目選擇測量項目(Item Selection)定量(quantitative)的 牽涉構念的不同構面(Dimensions)和關係網(nomological network)189 undergraduate studentsExploratory Factor Analysis(Items with high loadings):Chose 4 items for each dimension(total=16 items)Internal Consistency Reliability(.83 to.90)Correlations with IQ test,Life satisfaction,PowerlessnessCorrelations 的結果的結果-The EI dimensions were all mildly correlated(ranging from r=.13 to.42),which indicated that they were related but not identical dimensions.-All EI dimensions correlated significantly with life satisfaction.The correlation ranged from.16 to.46.-All EI dimensions correlated moderately and negatively with the powerlessness measures.The correlation ranged from-.13 to-.39.-Finally,the EI dimensions had minimal correlations with the IQ estimate.測量工具的效度檢定測量工具的效度檢定(Construct Validation)-172 undergraduate student sample-With the specified four items loading on their respective EI dimensions,CFA:(1)2=132.41(d.f.=98);(2)Standardized RMR=.08,CFI=.95;TLI=.93.-Negatively correlated with powerlessness-Positively correlated with life satisfaction測量工具的效度檢定測量工具的效度檢定(Construct Validation)-2146 undergraduate student sample-With the specified four items loading on their respective EI dimensions,CFA:(1)2=179.33(d.f.=98);(2)Standardized RMR=.07,CFI=.91;TLI=.89.-Negatively correlated with powerlessness-Positively correlated with life satisfaction測量工具的效度檢定測量工具的效度檢定(Construct Validation)-3110 undergraduate student sampleEFA together with Big-Five personality dimensionsTMM(another EI measure)After controlling for Big-Five and TMM,R2=.077(p.01)in predicting life satisfactionAfter controlling for Big-Five and TMM,R2=.059(p.10)in predicting powerlessness測量工具的效度檢定測量工具的效度檢定(Construct Validation)-3116 non-teaching employee sample in universityTwo measures of Big-Five dimensions and EICFA(3 indicators or each Big-Five dimension and 4 for EI):2=591.59(d.f.=398);Standardized RMR=.08;CFI=.90;TLI=.89.After controlling for Big-Five and EQ-i(another EI measure),R2=.023(p.10)in predicting job satisfactionMTMM他評的他評的EI測量工具測量工具-1444 Parent-Student sample in PRCBoth parent and student rated students Big-Five Personality and EI;Students also rated their own,Life satisfaction and PowerlessnessAfter controlling for Big-Five and demographics,parents ratings of student EI can predict life satisfaction(R2=.02,p.01)and powerlessness(R2=.02,p.01)MTMM of Parent-Student SampleCFA on the MTMM(EI and Big-Five;3 indicators for each trait)Two-method-six trait model is better than two method only,and six trait only modelSix traits explain 41.49%observed variance;two method explained 2226%observed variance;36.26%observed variance attributed to random error他評的他評的EI測量工具測量工具-2165 Supervisor-Worker-Colleague sample in PRCSupervisor,worker,and one colleague rated the workers job performance and EI;Supervisor rated the workers performanceAfter controlling for demographics and other work-related constructs(e.g.,LMX),colleagues ratings of EI can predict task performance(R2=.17,p.01),interpersonal facilitation(R2=.18,p.01),and job dedication(R2=.24,p.01)MTMM of Supervisor-Worker-Colleague SampleCFA on the MTMM(EI and 3 performance dimensions rated by three raters;3 indicators for each trait)Three-method-four trait model is better than three method only,and four trait only modelFour traits explain 25%observed variance;three method explained 70%observed variance;5%observed variance attributed to random error