An Approach to PCA-Principal Component Analysis (Analisis Komponen Utama)
Terbahagi kepada 2:
1. EFA
2. CFA
Take note:
Ø VARIMAX-assume the factors are not correlated
Ø DIRECT OBLIMIN- they are correlated
Ø Standard deviation ~ signaled which items are NO GOOD
Ø If using 7 point Likert Scale, SD BELOW 1 (not reliable to distinguish-cannot separate people easily)
Ø If using 5 point Likert Scale, SD 0.85 and above (reliable to distinguish-can separate people easily)
Ø SUMMATive score compared to FACTOR score- FACTOR score better
Ø REASONS FOR USING PCA
Data summarization
l identify representative variable(s)
l explore patterns of relationship(s)
l examine underlying dimensions/factors/ latent variables/constructs
l define the construct
Data reduction
l create a new set of variables
l use a new set of values
l use composite scores; interval variables
Construct Validation
· To provide empirical evidence that the data (responses) effectively and accurately measure a given psychological construct
Ø POSSIBLE RESEARCH QUESTIONS
l What are the factors that influence the respondents’ perceptions?
l Are the respondents’ perceptions a multi-dimensional construct?
l How many latent variables underlie the respondents’ perceptions?
l What are the underlying dimensions that account for variability of the respondents’ perceptions?
l Do the results of the hypothesized measurement model replicate those exploratory work which were conducted earlier?
Ø AN APPROACH TO APPLYING PCA
- Develop a measurement model theoretically
- Construct research questions
- Construct instrument
- Select sample à collect data
- Run item analysis- cronbach alpha
S5: Item Analysis-SCALE-RELIABILITY ANALYSIS-MASUKKAN ITEM YANG NAK ANALYSE-KLIK ITEM, SCALE, SCALE IF ITEM DELETED
l Std deviation: examine items with very
low SD (SD < .1)
l Alpha if item deleted
l Overall reliability index
l Sub-construct reliability index
l define population
l random sample
l adequate sample
l minimum: 5 respondents per item
l maximum: 15 respondents per item
l a sample for cross-validation
- Run PCA- factor analysis
7. S6: Run PCA : DATA REDUCTION-FACTOR-MASUKKAN ITEM YANG NAK DIBUAT FAKTOR ANALISIS-klik univariate-initial solution-koeficient-KMO-klik direct oblimin-rotated solution-save as variable-regression-
b. KMO, Bartlett’s Test
c. Anti-image matrix
d. Communalities matrix
e. PVE
f. Pattern Matrix
8. Evaluate the model fit
- Assessment of assumption : CORRELATED ITEMS
l Bartlett test of sphericity
l Overall measure of sampling adequacy (MSA) – MORE THAN 0.5 IS OK
 |
| MSA more than 0.5 is OK |
l Individual item MSA
- Number of extracted factors (components)
- Proportion of variance explained
- Number of items for each factor
- Direction and magnitude of factor loading
- Noises & contamination?
- Interpretability of each factor
- Reliability index for each factor
- Run validation analysis
- Present the results
Ø fieldwork + data analysis
Ø analyzing & interpreting the results
Ø writing the report
± 60% PVE, higher % better!
Variance Explained?
TOTAL VARIANCE EXPLAINED-KENA BENTANG/TERANGKAN DALAM BENTUK AYAT
lA priori criterion
l Eigenvalue for each component ≥ 1.00
l Scree test/plot criterion
