suggests that common method biases

suggests that common method biases do not explain relationships between study constructs. The CFA output was used to calculate the composite reliability (minimum 0.77) and average variance extracted (minimum 0.53) for each construct. Discriminant validity was assessed in two ways. First, we used a χ2 difference test for each possible pair of constructs, forcing each pair of constructs to fit a single-factor model and comparing the fit with a two-factor model (Anderson and Gerbing, 1988). Even accounting for the large number of χ2 tests performed (see Vorhees et al., 2016), the two-factor model always provided a better fit with the data than the single-factor model. Second, we compared the average variances extracted (AVEs) with the squared correlations from the standardized PHI matrix. The lowest AVE was 0.53 (market dynamism) and the largest squared correlation between any two constructs was 0.21, indicating good discriminant validity (Fornell and Larcker, 1981).

Structural model

The second stage of the analysis involved running the structural model with instrumental variables. Our approach here follows the recommendations of Venkatraman (1989) in analyzing fit-as-moderation relationships. Specifically, we eschew sub-group analyses or split sample approaches in favor of a moderated structural equation model because the performance outcome is determined by the interactions between the predictor and the moderators (Sharma et al., 1981; Venkatraman, 1989).

We mean-centered the raw scores of antecedent variables to reduce potential problems of multicollinearity linked to the inclusion of the interaction terms (Aiken and West, 1991) required for the assessment of moderating effects. Three interaction terms were created by the products of spontaneity with: strategic planning; centralization, and; market dynamism. In addition, the latter moderating variables were also inserted into the structural equations as main effects following statistical convention for hierarchical testing of interaction effects (Sharma et al., 1981). In line with Germann et al. (2013), we also computed quadratic terms (both for the main effect of spontaneity and for the moderating effects), and included them in the model to control for potential non-linear effects. We used Ping’s (1995) approach for estimating interactions between latent constructs in structural equation models. This procedure is recommended in order to lessen model complexity since our model comprised a number of interaction effects ( Jaccard and Wan, 1996). Single indicants were therefore computed for all multi-item latent variables (except for export profit effectiveness) by averaging the corresponding measurement items. Export profit effectiveness was modeled as a first-order latent variable comprised of three items. We set the error variances of the single indicants associated with the latent variables to [(1–α).σ2] ( Jöreskog and Sörbom, 1993), where α corresponds to the construct reliability and σ to the standard deviation of the single indicant. Following established guidelines (Song et al., 2005) we used the factor loading and the error variance estimates obtained from the main effects model to compute loadings and error variances of the single indicants corresponding to the quadratic and interaction terms. We ran two models, a model where endogeneity is assumed not to exist and a model where endogeneity is presumed and controlled for. The χ2 difference between those two models was not statistically significant, suggesting that endogeneity is not a concern (Antonakis et al., 2010).

In addition, we ran two models, namely, a constrained model and an unconstrained model. In the constrained model we allowed only the direct effects to be estimated freely. Accordingly, we set interaction terms at zero. In the unconstrained model

Results (Indonesian) 1:

menunjukkan bahwa bias metode umum tidak menjelaskan hubungan antara studi konstruksi. CFA output yang digunakan untuk menghitung komposit kehandalan (minimal 0.77) dan rata-rata varians diekstrak (minimal 0,53) untuk membangun masing-masing. Validitas diskriminan dinilai dalam dua cara. Pertama, kami menggunakan tes perbedaan χ2 untuk setiap pasangan mungkin konstruksi, memaksa setiap pasangan konstruksi yang cocok dengan model satu-faktor dan membandingkan yang cocok dengan model dua faktor (Anderson dan Gerbing, 1988). Bahkan akuntansi untuk sejumlah besar χ2 tes dilakukan (Lihat Vorhees et al., 2016), model dua faktor selalu disediakan yang lebih cocok dengan data dari model tunggal-faktor. Kedua, kita dibandingkan rata-rata varians diekstrak (AVEs) dengan korelasi kuadrat dari matriks PHI standar. AVE terendah 0,53 (dinamika pasar) dan kuadrat terbesar korelasi antara konstruksi setiap dua 0,21, menunjukkan baik diskriminan validitas (Fornell dan Larcker, 1981).Model strukturalTahap kedua dari analisis terlibat menjalankan model struktural dengan variabel instrumental. Pendekatan kami di sini mengikuti rekomendasi Venkatraman (1989) dalam menganalisis hubungan cocok sebagai moderasi. Secara khusus, kita menjauhkan diri analisis sub kelompok atau split sampel pendekatan mendukung model persamaan struktural dikelola karena hasil kinerja ditentukan oleh interaksi antara peramal dan moderator (Sharma et al., 1981; Venkatraman, 1989).We mean-centered the raw scores of antecedent variables to reduce potential problems of multicollinearity linked to the inclusion of the interaction terms (Aiken and West, 1991) required for the assessment of moderating effects. Three interaction terms were created by the products of spontaneity with: strategic planning; centralization, and; market dynamism. In addition, the latter moderating variables were also inserted into the structural equations as main effects following statistical convention for hierarchical testing of interaction effects (Sharma et al., 1981). In line with Germann et al. (2013), we also computed quadratic terms (both for the main effect of spontaneity and for the moderating effects), and included them in the model to control for potential non-linear effects. We used Ping’s (1995) approach for estimating interactions between latent constructs in structural equation models. This procedure is recommended in order to lessen model complexity since our model comprised a number of interaction effects ( Jaccard and Wan, 1996). Single indicants were therefore computed for all multi-item latent variables (except for export profit effectiveness) by averaging the corresponding measurement items. Export profit effectiveness was modeled as a first-order latent variable comprised of three items. We set the error variances of the single indicants associated with the latent variables to [(1–α).σ2] ( Jöreskog and Sörbom, 1993), where α corresponds to the construct reliability and σ to the standard deviation of the single indicant. Following established guidelines (Song et al., 2005) we used the factor loading and the error variance estimates obtained from the main effects model to compute loadings and error variances of the single indicants corresponding to the quadratic and interaction terms. We ran two models, a model where endogeneity is assumed not to exist and a model where endogeneity is presumed and controlled for. The χ2 difference between those two models was not statistically significant, suggesting that endogeneity is not a concern (Antonakis et al., 2010).In addition, we ran two models, namely, a constrained model and an unconstrained model. In the constrained model we allowed only the direct effects to be estimated freely. Accordingly, we set interaction terms at zero. In the unconstrained model

Results (Indonesian) 2:

menunjukkan bahwa metode umum bias tidak menjelaskan hubungan antara konstruk penelitian. CFA Output digunakan untuk menghitung reliabilitas komposit (minimum 0.77) dan rata-rata varians diekstraksi (minimum 0.53) untuk setiap konstruk. Validitas diskriminan dinilai dalam dua cara. Pertama, kami menggunakan uji beda χ2 untuk setiap pasangan yang mungkin dari konstruksi, memaksa setiap pasangan konstruksi sesuai model faktor tunggal dan membandingkan cocok dengan model dua faktor (Anderson dan Gerbing, 1988). Bahkan akuntansi untuk sejumlah besar tes χ2 dilakukan (lihat Vorhees et al., 2016), model dua faktor selalu tersedia lebih cocok dengan data dari model faktor tunggal. Kedua, kami membandingkan varians rata diekstraksi (aves) dengan korelasi kuadrat dari matriks PHI standar. The AVE terendah adalah 0,53 (pasar dinamika) dan korelasi kuadrat terbesar antara dua konstruk adalah 0,21, menunjukkan baik validitas diskriminan (Fornell dan Larcker, 1981).

Model Struktural

Tahap kedua dari analisis yang terlibat menjalankan model struktural dengan variabel instrumental. Pendekatan kami di sini mengikuti rekomendasi dari Venkatraman (1989) dalam menganalisis hubungan fit-sebagai-moderasi. Analisis khusus, kami menghindari sub-kelompok atau sampel perpecahan pendekatan demi sebuah model persamaan struktural dimoderasi karena hasil kinerja ditentukan oleh interaksi antara prediktor dan moderator (Sharma et al, 1981;. Venkatraman, 1989).

Kami berarti -centered skor baku dari variabel anteseden untuk mengurangi potensi masalah multikolinearitas terkait dengan masuknya istilah interaksi (Aiken dan Barat, 1991) diperlukan untuk penilaian efek moderasi. Tiga hal interaksi diciptakan oleh produk spontanitas dengan: perencanaan strategis; sentralisasi, dan; dinamisme pasar. Selain itu, variabel moderator yang terakhir juga dimasukkan ke dalam persamaan struktural sebagai efek utama mengikuti konvensi statistik untuk pengujian hirarkis efek interaksi (Sharma et al., 1981). Sejalan dengan Germann et al. (2013), kami juga dihitung dari segi kuadrat (baik untuk efek utama spontanitas dan untuk efek moderasi), dan termasuk dalam model untuk mengontrol efek non-linear potensial. Kami menggunakan (1995) pendekatan Ping untuk memperkirakan interaksi antara konstruk laten dalam model persamaan struktural. Prosedur ini dianjurkan untuk mengurangi kerumitan model karena model kami terdiri sejumlah efek interaksi (Jaccard dan Wan, 1996). Oleh karena itu indicants tunggal dihitung untuk semua variabel laten multi-item (kecuali untuk efektivitas keuntungan ekspor) dengan rata-rata item pengukuran yang sesuai. Efektivitas laba ekspor dimodelkan sebagai orde pertama variabel laten terdiri dari tiga item. Kami mengatur varians kesalahan dari indicants tunggal yang terkait dengan variabel laten untuk [(1-α) .σ2] (Joreskog dan Sörbom, 1993), di mana α sesuai dengan kehandalan membangun dan σ untuk deviasi standar dari indicant tunggal. Berikut pedoman yang ditetapkan (Lagu et al., 2005) kami menggunakan factor loading dan perkiraan kesalahan varians diperoleh dari model efek utama untuk menghitung beban dan varians kesalahan dari indicants tunggal sesuai dengan persyaratan kuadrat dan interaksi. Kami berlari dua model, model mana endogeneity diasumsikan tidak ada dan model mana endogeneity dianggap dan dikendalikan untuk. Perbedaan χ2 antara dua model secara statistik tidak signifikan, menunjukkan endogeneity yang tidak perhatian (Antonakis et al., 2010).

Selain itu, kami berlari dua model, yaitu, model dibatasi dan model tidak dibatasi. Dalam model dibatasi kita diperbolehkan hanya efek langsung diperkirakan bebas. Dengan demikian, kami menetapkan hal interaksi nol. Dalam model tidak dibatasi