Aggregation and Analyses of Effect

Aggregation and Analyses of Effect Sizes
In our analyses, a number of studies were published on the same longitudinal sample sets. To balance the optimization of the contribution of reported results, with the sample attenuation required for sample independence, we adopted the procedure used in recent meta-analyses of trait consistency (Roberts & DelVecchio, 2000; Trzesniewski et al., 2003). Stability coefficients were aggregated by sample rather than by study.
We created aggregated databases from the overall database to test the moderating effects of age and other variables on interest stability. To test the relation between interest stability and age, we aggregated sample data within the age categories by age at the initiation of the study. For studies that reported results based on multiple waves of testing, we included only the results from nonoverlapping time intervals. For example, Strong (1951) reported six stability coefficients for ages 22–27, 22–32, 22–44, 27–32, 27–44, and 32–44, based on four times of testing (i.e., ages 22, 27, 32, and 44) of a single sample. On the basis of the above criterion, we used only three of the six coefficients, each of which represents a corresponding wave of the three waves of testing (i.e., ages 22–27, 27–32, and 32–44). If, over the same time period, stability for a sample was represented by both rank-order and profile correlations or if stability was indexed at more than one scale-generality level for the same sample, we averaged the coefficients into the age category that was represented when the initial assessment of the particular sample took place. This technique meant that each longitudinal sample could contribute an averaged coefficient to several age categories, but each sample would contribute only once to the aggregated estimate for each age period.
To compute the estimates of interest stability, we followed Hedges and Olkin’s (1985) recommendations (also see Roberts & DelVecchio, 2000; Trzesniewski et al., 2003). The effect size estimates consisted of Fisher’s Z-transformed correlation coefficients, which we then weighted by the inverse of the variance when making population estimates. We obtained the estimated population correlations () through a Z-to-r transformation of the effect size estimates. We calculated confidence intervals (CIs) and tests of heterogeneity using the formulas provided by Hedges and Olkin (1985). Using this procedure, we obtained effect size estimates for each of the eight age categories.
Similar aggregation and calculation procedures were performed on databases that were aggregated by cohort standing, type of stability coefficient, test–retest interval, gender, generality of scales, and interest classification. Similar to the age-based database, the sample data were first aggregated by potential moderator, such as gender, and then by sample. In other words, each sample could only contribute one averaged estimate of interest stability to each category within each moderator.
Although the reaggregation of samples by proposed moderators provided the means of examining the main effects these variables might have on effect size estimates, the procedure did not test for moderating effects. For instance, if there were significant differences among the categories in the database that was aggregated by gender, we could conclude that men’s interests were, for instance, more stable than women’s interests. However, a conclusion about the effects of gender differences on interest stability across the age categories cannot be made, because gender was not considered in the derivation of the age-based population estimates.

Results (Indonesian) 1:

Agregasi dan analisis ukuran efekDalam analisis kami, sejumlah studi yang diterbitkan pada set longitudinal sampel yang sama. Untuk menyeimbangkan optimasi kontribusi melaporkan hasil, dengan redaman sampel yang diperlukan untuk kemerdekaan sampel, kami mengadopsi prosedur yang digunakan dalam meta-analisis sifat konsistensi (Roberts & DelVecchio, 2000; Trzesniewski et al., 2003). Koefisien stabilitas yang dikumpulkan oleh sampel bukan oleh studi.Kami menciptakan database gabungan dari database secara keseluruhan untuk menguji efek pengaruh umur dan variabel lain bunga stabilitas. Untuk menguji hubungan antara bunga stabilitas dan usia, kita dikumpulkan data sampel dalam kategori umur usia di inisiasi dari studi. Studi yang melaporkan hasil berdasarkan beberapa gelombang pengujian, kami menyertakan hanya hasil dari interval waktu nonoverlapping. Misalnya, kuat (1951) melaporkan enam stabilitas koefisien untuk usia 22-27, 22-32, 22-44, 27-32, 27-44 dan 32-44, berdasarkan empat kali dari pengujian (yakni, usia 22, 27, 32, dan 44) sampel tunggal. Berdasarkan kriteria di atas, kami menggunakan hanya tiga dari koefisien enam, yang masing-masing mewakili gelombang yang sesuai dari tiga gelombang pengujian (yakni, usia 22-27, 27-32 dan 32-44). Jika, selama periode waktu yang sama, stabilitas untuk sampel diwakili oleh korelasi urutan peringkat dan profil atau jika stabilitas diindeks di lebih dari satu tingkat skala-keumuman untuk sampel yang sama, kami rata-rata koefisien ke dalam kategori umur diwakili ketika penilaian awal sampel tertentu terjadi. Teknik ini dimaksudkan bahwa setiap sampel longitudinal bisa berkontribusi koefisien rata-rata untuk beberapa kategori umur, tapi setiap sampel akan hanya sekali berkontribusi agregat perkiraan untuk setiap zaman.To compute the estimates of interest stability, we followed Hedges and Olkin’s (1985) recommendations (also see Roberts & DelVecchio, 2000; Trzesniewski et al., 2003). The effect size estimates consisted of Fisher’s Z-transformed correlation coefficients, which we then weighted by the inverse of the variance when making population estimates. We obtained the estimated population correlations () through a Z-to-r transformation of the effect size estimates. We calculated confidence intervals (CIs) and tests of heterogeneity using the formulas provided by Hedges and Olkin (1985). Using this procedure, we obtained effect size estimates for each of the eight age categories.Similar aggregation and calculation procedures were performed on databases that were aggregated by cohort standing, type of stability coefficient, test–retest interval, gender, generality of scales, and interest classification. Similar to the age-based database, the sample data were first aggregated by potential moderator, such as gender, and then by sample. In other words, each sample could only contribute one averaged estimate of interest stability to each category within each moderator.Although the reaggregation of samples by proposed moderators provided the means of examining the main effects these variables might have on effect size estimates, the procedure did not test for moderating effects. For instance, if there were significant differences among the categories in the database that was aggregated by gender, we could conclude that men’s interests were, for instance, more stable than women’s interests. However, a conclusion about the effects of gender differences on interest stability across the age categories cannot be made, because gender was not considered in the derivation of the age-based population estimates.