Tulisan berasal dari:
Jorg Henseler, Christian M. Ringle and Rudolf R. Sinkovics (2009),
The Use Of Partial Least Squares Path Modeling In International Marketing, New Challenges to International Marketing Advances in International Marketing, Volume 20, 277–319
---------------------------------------------------------------------------------------------
PLS path models are formally defined by two sets of linear equations: the inner model and the outer model. The inner model specifies the relationships between unobserved or latent variables, whereas the outer model specifies the relationships between a latent variable and its observed or manifest variables. The various literatures do not always employ the same terminology. For instance, publications addressing CBSEM (e.g., Rigdon, 1998) often refer to structural models and measurement models or (observed) indicator variables; whereas those focusing on PLS path modeling (e.g., Lohmo¨ller, 1989) use the terms inner model and outer model or manifest variables for similar elements of the causal model.
--------------------------------------------------------------------------------------------
Model SEM PLS secara formal terdiri atas 2 set persamaan linier yaitu inner model dan outer model. Inner model menspesifikasikan hubungan antara variabel laten dengan variable manifest atau indikator sedangkan model outer menspesifikasikan hubungan antara variabel laten dengan variabe laten lainnya. Beberapa literatur tidak menggunakan terminologi inner model dan outer model akan tetapi model pengukuran dan model struktural. Akan tetapi dalam model PLS ini disebutkan inner model yang mempunyai makna sama dengan model struktural dan outer model yang sama dengan model struktural.
------------------------------------------------------------------------------------------
Structural equation models usually involve latent variables with multiple indicators. The measurement model or outer model specifies the relationship between indicators and latent variables. The direction of path relationships per measurement model and, thus, the causality between the latent variable and its indicators are either described by a reflective or a formative mode. The reflective measurement model has its roots in classical test theory and psychometrics (Nunnally & Bernstein, 1994). Each indicator represents an error-afflicted measurement of the latent variable. The direction of causality is from the construct to the indicators; thus, observed measures are assumed to reflect variation in the latent variable. In other words, changes in the construct are expected to be manifested in changes in all of its indicators
-----------------------------------------------------------------------------------------
Model pengukuran ada 2 (dua) jenis yaitu model pengukuran reflektif dan formatif. Model pengukuran reflektif berakar dari teori pengukuran klasik. Setiap indikator mewakili pengukuran variabel laten. Arah kausalitas mengalir dari variable laten/ konstrak ke indikator pengukur. indikator atau indikator pengukur diasumsikan menggambarkan varians dalam variabe laten. Setiap perubahan dalam variabe laten maka diharapkan akan memberikan perubahan pada indikatornya.
-----------------------------------------------------------------------------------------
Although the inclusion of formative measures in CBSEM has been well documented (e.g., Jo¨reskog & Goldberger, 1975; MacCallum & Browne, 1993; Jo¨reskog & Sorbom, 1996), analysts usually encounter identification problems. As a sort of ad-hoc remedy, formative indicators could be modeled in CBSEM by re-specifying the formative indicators as exogenous latent variables with single indicators, fixed unit loadings, and a fixed measurement error (Williams, Edwards, & Vandenberg, 2003). In contrast, similar problems do not arise in PLS path modeling. The PLS path modeling algorithm – is equally well suited for SEM with reflective and/or formative measurement models. The only problematic issue, however, is connected to manifest variables’ critical level of multicollinearity in formative measurement models.
-----------------------------------------------------------------------------------------
Model pengukuran formatif dalam SEM Kovarians menghadapi permasalahan identifikasi. Meskipun beberapa referensi para ahli model pengukuran formatif masih dapat dilakukan dalam SEM Kovarians dengan menspesifikasikan ulang indikator formative sebagai variabel laten tunggal, memberikan bobot tetap pada nilai loadingnya (fixed unit loading) dan error pengukuran tetap. Problem ini tidak terjadi dalam SEM PLS. Algoritma dalam SEM PLS mengakomodasi untuk model pengukuran reflektif dan formatif. hanya saja permasalahan model formatif dalam SEM PLS adalah terkait multikolinieritas.
-----------------------------------------------------------------------------------------
PLS path modeling does not provide any global goodness-of-fit criterion. As a consequence, Chin (1998) has put forward a catalog of criteria to assess partial model structures. A systematic application of these criteria is a two-step process, encompassing (1) the assessment of the outer model and (2) the assessment of the inner model. At the beginning of the two step process, model assessment focuses on the measurement models. A systematic evaluation of PLS estimates reveals the measurement reliability and validity according to certain criteria that are associated with formatve and reflective outer model. It only makes sense to evaluate the inner path model estimates when the calculated latent variable scores show evidence of sufficient reliability and validity.
-----------------------------------------------------------------------------------------
Model SEM PLS tidakempunyai kecocokan model secara menyeluruh (goodness of fit). sebagai akibatnya maka Chin menyarankan untuk mengevaluasi model PLS dalam 2 (dua) tahap: 1) menguji atau evaluasi model pengukuran dan 2) evaluasi model struktural. Untuk model pengukuran memberitahukan informasi reliabilitas dan validitas sesuai dengan kriteria tertentu baik pengukuran reflektif atau formatif.
----------------------------------------------------------------------------------------
Reflective measurement models should be assessed with regard to their reliability and validity. Usually, the first criterion which is checked is internal consistency reliability. The traditional criterion for internal consistency is Cronbach’s a (Cronbach, 1951), which provides an estimate for the reliability based on the indicator intercorrelations. While Cronbach’s a assumes that all indicators are equally reliable, PLS prioritizes indicators according to their reliability, resulting in a more reliable composite. As Cronbach’s a tends to provide a severe underestimation of the internal consistency reliability of latent variables in PLS path models, it is more appropriate to apply a different measure, the composite reliability rc (Werts, Linn, & Jo¨reskog, 1974). The composite reliability takes into account that indicators have different loadings, and can be interpreted in the same way as Cronbach’s a. No matter which particular reliability coefficient is used, an internal consistency reliability value above 0.7 in early stages of research and values above 0.8 or 0.9 in more advanced stages of research are regarded as satisfactory (Nunnally & Bernstein, 1994), whereas a value below 0.6 indicates a lack of reliability.
----------------------------------------------------------------------------------------------
Untuk model pengukuran reflektif maka memerlukan pemeriksaan reliabilitas dan validitas. Kriteria pertama adalah reliabilitas konsistensi internal. metode umumnya yang digunakan adalah Cronbach's Alpha. Ukuran ini mengasumsikan bahwa semua indikator mempunyai tingkat keandalan/reliable/bobot yang sama sehingga menghasilkan taksiran reliabilitas yang yang kurang tepat. Oleh karena itu maka disarankan mempunytai ukuran yang lainnya adalah composite reliability (CR) yang dikembangkan oleh (Werts. Linn. & Joreskog, 1974). CR mengukur tingkat reliabilitas berdasarkan loadings yang berbeda dan dapat diinterpretasikan sama seperti Cronbach's Alpha. Nilai CR yang diharapkan adalah lebih dari 0,70 untuk studi awal atau 0.80 dan 0.90 untuk studi lanjutan. Sedangkan nilai CR dibawah 0.60 menunjukan tingkat reliabilitas yang lemah.
----------------------------------------------------------------------------------------------
As the reliability of indicators varies, the reliability of each indicator should be assessed. Researchers postulate that a latent variable should explain a substantial part of each indicator’s variance (usually at least 50%). Accordingly, the absolute correlations between a construct and each of its manifest variables (i.e. the absolute standardized outer loadings) should be p higher than 0.7 ( 0.5). Moreover, some psychometrists (e.g., Churchill, 1979) recommend eliminating reflective indicators from measurement models if their outer standardized loadings are smaller than 0.4. Taking into account PLS’ characteristic of consistency at large, one should be careful when eliminating indicators. Only if an indicator’s reliability is low and eliminating this indicator goes along with a substantial increase of composite reliability, it makes sense to discard this indicator.
----------------------------------------------------------------------------------------------
Pemeriksaan reliabilitas setiap indikator perlu dilakukan dengan melihat standardized loadings. Nilai minimal yang diharapkan adalah 0.70 atau 0.50. saran dari Churchill (1979) adalah nilai loading kurang dari 0.40 dihlangkan dalam model pengukuran dengan ketentuan adalah menghilangkan sebuah indikator yang rendah dan meningkatkan secara substansi nilai komposit reliabilitas.
----------------------------------------------------------------------------------------------
For the assessment of validity, two validity subtypes are usually examined: the convergent validity and the discriminant validity. Convergent validity signifies that a set of indicators represents one and the same underlying construct, which can be demonstrated through their unidimensionality. Fornell and Larcker (1981) suggest using the average variance extracted (AVE) as a criterion of convergent validity. An AVE value of at least 0.5 indicates sufficient convergent validity, meaning that a latent variable is able to explain more than half of the variance of its indicators on average (e.g., Go¨tz, Liehr-Gobbers, & Krafft, 2009). Discriminant validity is a rather complementary concept: Two conceptually different concepts should exhibit sufficient difference (i.e. the joint set of indicators is expected not to be unidimensional). In PLS path modeling, two measures of discriminant validity have been put forward: The Fornell–Larcker criterion and the cross-loadings. The Fornell–Larcker criterion (Fornell & Larcker, 1981) postulates that a latent variable shares more variance with its assigned indicators than with any other latent variable
---------------------------------------------------------------------------------------------
Konvergen validitas menjelaskan sejumlah indikator mengukur konsep atau variable yang sama (variable laten). Validitas konvergen terlihat dari tingginya korelasi antara indikator dalam mengukur variable laten tersebut. Konvergen validity terlihat dari nilai AVE dengan nilai minimum yang diharapkan adalah 0.50 yang berarti variable laten mampu menjelaskan 50% atau setengah variasi indikatornya secara rata-rata. Ukuran validitas lainnya yaitu validitas discriminan yang merupakan pelengkap dari validitas konvergen. Ukuran ini ditunjukan oleh rendahnya korelasi antara indikator yang mengukur variable laten yang berbeda, Menurut Fornelll-Lacker kriteria yang digunakan adalah cross loading.
PLS path modeling does not provide any global goodness-of-fit criterion. As a consequence, Chin (1998) has put forward a catalog of criteria to assess partial model structures. A systematic application of these criteria is a two-step process, encompassing (1) the assessment of the outer model and (2) the assessment of the inner model. At the beginning of the two step process, model assessment focuses on the measurement models. A systematic evaluation of PLS estimates reveals the measurement reliability and validity according to certain criteria that are associated with formatve and reflective outer model. It only makes sense to evaluate the inner path model estimates when the calculated latent variable scores show evidence of sufficient reliability and validity.
-----------------------------------------------------------------------------------------
Model SEM PLS tidakempunyai kecocokan model secara menyeluruh (goodness of fit). sebagai akibatnya maka Chin menyarankan untuk mengevaluasi model PLS dalam 2 (dua) tahap: 1) menguji atau evaluasi model pengukuran dan 2) evaluasi model struktural. Untuk model pengukuran memberitahukan informasi reliabilitas dan validitas sesuai dengan kriteria tertentu baik pengukuran reflektif atau formatif.
----------------------------------------------------------------------------------------
Reflective measurement models should be assessed with regard to their reliability and validity. Usually, the first criterion which is checked is internal consistency reliability. The traditional criterion for internal consistency is Cronbach’s a (Cronbach, 1951), which provides an estimate for the reliability based on the indicator intercorrelations. While Cronbach’s a assumes that all indicators are equally reliable, PLS prioritizes indicators according to their reliability, resulting in a more reliable composite. As Cronbach’s a tends to provide a severe underestimation of the internal consistency reliability of latent variables in PLS path models, it is more appropriate to apply a different measure, the composite reliability rc (Werts, Linn, & Jo¨reskog, 1974). The composite reliability takes into account that indicators have different loadings, and can be interpreted in the same way as Cronbach’s a. No matter which particular reliability coefficient is used, an internal consistency reliability value above 0.7 in early stages of research and values above 0.8 or 0.9 in more advanced stages of research are regarded as satisfactory (Nunnally & Bernstein, 1994), whereas a value below 0.6 indicates a lack of reliability.
----------------------------------------------------------------------------------------------
Untuk model pengukuran reflektif maka memerlukan pemeriksaan reliabilitas dan validitas. Kriteria pertama adalah reliabilitas konsistensi internal. metode umumnya yang digunakan adalah Cronbach's Alpha. Ukuran ini mengasumsikan bahwa semua indikator mempunyai tingkat keandalan/reliable/bobot yang sama sehingga menghasilkan taksiran reliabilitas yang yang kurang tepat. Oleh karena itu maka disarankan mempunytai ukuran yang lainnya adalah composite reliability (CR) yang dikembangkan oleh (Werts. Linn. & Joreskog, 1974). CR mengukur tingkat reliabilitas berdasarkan loadings yang berbeda dan dapat diinterpretasikan sama seperti Cronbach's Alpha. Nilai CR yang diharapkan adalah lebih dari 0,70 untuk studi awal atau 0.80 dan 0.90 untuk studi lanjutan. Sedangkan nilai CR dibawah 0.60 menunjukan tingkat reliabilitas yang lemah.
----------------------------------------------------------------------------------------------
As the reliability of indicators varies, the reliability of each indicator should be assessed. Researchers postulate that a latent variable should explain a substantial part of each indicator’s variance (usually at least 50%). Accordingly, the absolute correlations between a construct and each of its manifest variables (i.e. the absolute standardized outer loadings) should be p higher than 0.7 ( 0.5). Moreover, some psychometrists (e.g., Churchill, 1979) recommend eliminating reflective indicators from measurement models if their outer standardized loadings are smaller than 0.4. Taking into account PLS’ characteristic of consistency at large, one should be careful when eliminating indicators. Only if an indicator’s reliability is low and eliminating this indicator goes along with a substantial increase of composite reliability, it makes sense to discard this indicator.
----------------------------------------------------------------------------------------------
Pemeriksaan reliabilitas setiap indikator perlu dilakukan dengan melihat standardized loadings. Nilai minimal yang diharapkan adalah 0.70 atau 0.50. saran dari Churchill (1979) adalah nilai loading kurang dari 0.40 dihlangkan dalam model pengukuran dengan ketentuan adalah menghilangkan sebuah indikator yang rendah dan meningkatkan secara substansi nilai komposit reliabilitas.
----------------------------------------------------------------------------------------------
For the assessment of validity, two validity subtypes are usually examined: the convergent validity and the discriminant validity. Convergent validity signifies that a set of indicators represents one and the same underlying construct, which can be demonstrated through their unidimensionality. Fornell and Larcker (1981) suggest using the average variance extracted (AVE) as a criterion of convergent validity. An AVE value of at least 0.5 indicates sufficient convergent validity, meaning that a latent variable is able to explain more than half of the variance of its indicators on average (e.g., Go¨tz, Liehr-Gobbers, & Krafft, 2009). Discriminant validity is a rather complementary concept: Two conceptually different concepts should exhibit sufficient difference (i.e. the joint set of indicators is expected not to be unidimensional). In PLS path modeling, two measures of discriminant validity have been put forward: The Fornell–Larcker criterion and the cross-loadings. The Fornell–Larcker criterion (Fornell & Larcker, 1981) postulates that a latent variable shares more variance with its assigned indicators than with any other latent variable
---------------------------------------------------------------------------------------------
Konvergen validitas menjelaskan sejumlah indikator mengukur konsep atau variable yang sama (variable laten). Validitas konvergen terlihat dari tingginya korelasi antara indikator dalam mengukur variable laten tersebut. Konvergen validity terlihat dari nilai AVE dengan nilai minimum yang diharapkan adalah 0.50 yang berarti variable laten mampu menjelaskan 50% atau setengah variasi indikatornya secara rata-rata. Ukuran validitas lainnya yaitu validitas discriminan yang merupakan pelengkap dari validitas konvergen. Ukuran ini ditunjukan oleh rendahnya korelasi antara indikator yang mengukur variable laten yang berbeda, Menurut Fornelll-Lacker kriteria yang digunakan adalah cross loading.
Salam,
Sofyan Yamin
1812 1825 2356
@ Bogor Desember 2017
Tidak ada komentar:
Posting Komentar