1 Introduksi
1.1 Latar Belakang
Replication dan center points adalah dua pillar DOE selain randomization (Fisher, 1935; Montgomery, 2017). Replication memberi pure error estimate ($\sigma^2$) yang independent dari model — esensial untuk inferensi. Center points ($x_i = 0$ semua faktor) memberi lack-of-fit (LOF) test + deteksi curvature — penting untuk validasi linear assumption sebelum lanjut ke quadratic RSM.
Replicate & Center Point Checker adalah modul diagnostik DOE post-execution yang mengevaluasi cukupkah replikasi + center points untuk inferensi yang valid, lalu menjalankan curvature test dan lack-of-fit test.
1.2 Tujuan Modul
Modul Replicate & Center Point Checker di SQalytics ditujukan untuk:
- Menerima executed design data (response + factor levels).
- Mengevaluasi replicate adequacy ($df_{\text{error}} \geq 8$ rekomendasi).
- Menjalankan curvature F-test dari center points.
- Menjalankan lack-of-fit F-test dari pure error vs lack-of-fit.
- Memberikan rekomendasi: lanjut analisis ANOVA, augment dengan axial points untuk RSM, atau redo dengan more replicates.
- Audiens: praktisi DOE setelah eksekusi run.
1.3 Posisi di Antara Alternatif
Pilih Replicate & Center Point Checker setelah eksekusi DOE sebelum ANOVA. Untuk design generation, pakai modul DOE generator. Untuk post-hoc ANOVA, pakai ANOVA Studio. Untuk augmentasi ke RSM, pakai RSM Studio.
2 Metode
2.1 Dasar Teoretis
Pure error dari replicates di same factor levels (mis. center points $n_C \geq 3$):
$$ \text{SS}_{\text{pure error}} = \sum_{c=1}^{n_C} (y_c - \bar{y}_C)^2 $$dengan $df_{\text{pure}} = n_C - 1$ (untuk center points alone) atau lebih bila ada replicate factorial points.
Curvature F-test (Box, Hunter, & Hunter, 2005, p. 251):
$F_{\text{curv}} \sim F_{1, n_C - 1}$. Bila $p_{\text{curv}} < 0.05$ → curvature signifikan → linear model insufficient → augment ke RSM (CCD axial points).
Lack-of-fit test (Kutner, Nachtsheim, Neter, & Li, 2005, ch. 6):
$$ F_{\text{LOF}} = \frac{\text{MS}_{\text{LOF}}}{\text{MS}_{\text{pure}}} $$dengan $\text{MS}_{\text{LOF}} = \text{SS}_{\text{LOF}} / (k - p)$ dan $\text{MS}_{\text{pure}} = \text{SS}_{\text{pure}} / (n - k)$. Bila $F_{\text{LOF}}$ signifikan → model tidak fit (perlu higher order terms).
Replicate adequacy untuk power 0.8 detect effect $\delta = 2\sigma$:
$$ n_{\text{rep}} \geq \frac{8 \sigma^2}{\delta^2 \cdot N_{\text{distinct}}} $$Rule of thumb: $df_{\text{error}} \geq 8$ untuk acceptable F-test power.
2.2 Persamaan Inti
Pure error SS: $\text{SS}_{\text{pure}} = \sum (y_c - \bar{y}_C)^2$
Curvature F-test: $F = n_F n_C (\bar{y}_F - \bar{y}_C)^2 / [(n_F + n_C) \text{MS}_{\text{pure}}]$
Lack-of-fit F-test: $F_{\text{LOF}} = \text{MS}_{\text{LOF}} / \text{MS}_{\text{pure}}$
Replicate adequacy: $df_{\text{error}} \geq 8$
2.3 Asumsi & Batas Validitas
| Asumsi | Konsekuensi jika dilanggar | Cara cek di SQalytics |
|---|---|---|
| Center points $\geq 3$ untuk curvature test | Test power rendah | Modul flag $n_C < 3$ |
| Replicate at identical factor levels | Pseudo-replication confounded | Cek levels exact match |
| Random order execution | Time bias di pure error | Cek run order |
| Same operator + day untuk center replicates | Block effect masuk pure error | Restricted block |
| Independent observations | Correlated errors | Time series check |
3 Cara Kerja
3.1 Step-by-Step di SQalytics
- Buka
Replicate & Center Point Checker. - Muat tabel: factor levels + response + run order.
- (Auto) Modul identifies center points ($x_i = 0$ all) dan replicates (same level combinations).
- Klik
Run Diagnostic. - Tinjau hasil: Tab
Replicate Summary— count, pure error, df; TabCurvature Test— F-stat, p-value, decision; TabLack-of-Fit Test— F-stat, p-value; TabRecommendation— augment, redo, atau proceed.
3.2 Template Tabel Input + Contoh Data Sintetis
Factorial $2^3$ + 4 center points:
| Run | T | t | C | Response |
|---|---|---|---|---|
| 1 | −1 | −1 | −1 | 12.5 |
| 2 | +1 | −1 | −1 | 18.2 |
| 3 | −1 | +1 | −1 | 15.4 |
| 4 | +1 | +1 | −1 | 22.1 |
| 5 | −1 | −1 | +1 | 14.8 |
| 6 | +1 | −1 | +1 | 19.5 |
| 7 | −1 | +1 | +1 | 17.0 |
| 8 | +1 | +1 | +1 | 23.4 |
| 9 | 0 | 0 | 0 | 24.5 (center) |
| 10 | 0 | 0 | 0 | 24.8 (center) |
| 11 | 0 | 0 | 0 | 24.2 (center) |
| 12 | 0 | 0 | 0 | 24.6 (center) |
3.3 Contoh Luaran
Replicate Summary:
| Metric | Value |
|---|---|
| Factorial points | 8 |
| Center points | 4 |
| Pure error SS (from center) | 0.20 |
| $df_{\text{pure}}$ | 3 |
| $\text{MS}_{\text{pure}}$ | 0.067 |
| $\bar{y}_F$ (factorial mean) | 17.86 |
| $\bar{y}_C$ (center mean) | 24.53 |
Curvature Test:
| Metric | Value |
|---|---|
| $F_{\text{curv}}$ | $8 \cdot 4 \cdot (24.53 - 17.86)^2 / (12 \cdot 0.067) = 1773$ |
| $p_{\text{curv}}$ | < 0.001 ✗ |
| Decision | Curvature signifikan — augment ke RSM (CCD) |
Replicate & Center Point Checker untuk $2^3$ factorial + 4 center points (data sintetis). (a) Scatter plot center points vs factorial means: 4 center point replicates ($\bar{y}_C = 24.53$, ditampilkan sebagai titik biru, run-number italic abu-abu di pojok kiri-atas tiap titik) vs 8 factorial corners ($\bar{y}_F = 17.86$, titik abu-abu); garis putus-putus = prediksi linear dari model; gap besar ($\Delta = 6.67$) menunjukkan curvature signifikan. (b) Bar chart F-test: bar kiri = $F_{\text{curv}} = 1773$ (merah, sangat tinggi) dan bar kanan = $F_{\text{critical}} = 10.13$ ($\alpha = 0.05$, garis referensi abu-abu); anotasi RE = $F_{\text{curv}} / F_{\text{critical}} = 174.9\times$ ditempatkan di ruang kosong antara dua bar tanpa overlapping; label "(augment ke RSM)" di atas bar kiri. Gradasi warna teal→merah pada bar intensitas menunjukkan severity level. Legend bottom-center di bawah kedua panel. Kesimpulan: linear model inadequate — augment ke CCD wajib (Box et al., 2005; Kutner et al., 2005).4 Kesimpulan
4.1 Relevansi Real-World
- Post-execution DOE checkpoint sebelum analisis ANOVA.
- Validation linear assumption untuk screening design.
- Trigger augmentation ke CCD / Box-Behnken ketika curvature significant.
- Pure error estimation untuk power study future experiments.
4.2 Where to Go from Here
⚙ Troubleshooting Cepat
i Riwayat Revisi
| Tanggal | Revisi | Penulis |
|---|---|---|
| 2026-05-12 | Migrasi MD v2 → HTML final dengan figure publikasi + caption Elsevier-style (W2 batch 12 Mei) | Claude |
| 2026-05-12 | Draft v2 publikasi (KaTeX curvature/LOF F-test + APA Box/Montgomery/Kutner) | Claude |
4 Referensi
- Box, G. E. P., Hunter, J. S., & Hunter, W. G. (2005). Statistics for experimenters: Design, innovation, and discovery (2nd ed.). John Wiley & Sons.
- Montgomery, D. C. (2017). Design and analysis of experiments (9th ed.). John Wiley & Sons.
- Kutner, M. H., Nachtsheim, C. J., Neter, J., & Li, W. (2005). Applied linear statistical models (5th ed.). McGraw-Hill/Irwin.
- Myers, R. H., Montgomery, D. C., & Anderson-Cook, C. M. (2016). Response surface methodology (4th ed.). John Wiley & Sons.