Other Software and Manual Methods of Analysis
❦
Choosing Software
| Software | Cost | Strengths | Typical use in business research |
|---|---|---|---|
| SPSS | Licensed (university licences vary) | Menu-driven; widely taught | Surveys: descriptives, tests, regression |
| Microsoft Excel | Licensed | Familiar; ToolPak for basic statistics; charts | Data entry, simple analyses, financial models |
| JASP and jamovi | Free | SPSS-like interface; Bayesian options | Free alternative to SPSS |
| R (with RStudio) | Free | Powerful and flexible; packages such as psych, lavaan, and ggplot2 | Advanced statistics, SEM, graphics, reproducible research |
| Python (pandas, statsmodels, scikit-learn) | Free | Data handling, machine learning, automation | Big data, text analysis, prediction |
| Stata and EViews | Licensed | Econometrics, panel data, time series | Finance and economics studies |
| AMOS | Licensed (with SPSS) | Covariance-based SEM with a graphical interface | CFA and SEM |
| SmartPLS | Licensed (free student version with limits) | PLS-SEM with a graphical interface | Models with many constructs, small samples |
| NVivo, ATLAS.ti, MAXQDA | Licensed | Qualitative coding, retrieval, and visualization | Interviews, focus groups, documents |
| Taguette, QualCoder | Free | Basic qualitative coding | Free alternatives |
| Zotero, Mendeley | Free | Reference management | Literature and citations |
Choose software that your university supports, that suits your analysis, and that you can learn in time. Ask your faculty about licences and training.
Structural Equation Modeling (SEM)
SEM tests models with latent variables and multiple relationships. There are two main families (Hair et al., 2019; Kline, 2016):
Covariance-based SEM (CB-SEM) using AMOS, Mplus, or R (lavaan): suitable for theory testing and larger samples.
1. Confirmatory factor analysis (CFA) for the measurement model.
2. Evaluate fit, commonly using: χ²/df below 3; CFI and TLI at or above 0.90 to 0.95; RMSEA at or below 0.06 to 0.08; SRMR at or below 0.08 (Hu & Bentler, 1999).
3. Assess convergent validity (standardized loadings above 0.50 to 0.70, average variance extracted (AVE) at or above 0.50, composite reliability at or above 0.70) and discriminant validity (the Fornell-Larcker criterion, Fornell & Larcker, 1981; or the HTMT ratio below 0.85 to 0.90, Henseler et al., 2015).
4. Test the structural model and hypotheses.
Partial least squares SEM (PLS-SEM) using SmartPLS: often chosen for prediction-oriented models, complex models, and smaller samples (Hair et al., 2019, 2022). Typical evaluation:
| Step | Criterion (common guidelines) |
|---|---|
| Indicator reliability | Outer loadings above 0.708 (0.40 to 0.70 may be retained if AVE and CR are met) |
| Internal consistency | Composite reliability between 0.70 and 0.95 |
| Convergent validity | AVE at or above 0.50 |
| Discriminant validity | HTMT below 0.90 (0.85 for conceptually distinct constructs); Fornell-Larcker |
| Collinearity | Inner VIF below 3 (below 5 acceptable) |
| Structural model | Bootstrapping (commonly 5 000 subsamples) for path significance; R² (about 0.25 weak, 0.50 moderate, 0.75 substantial); f² effect size (0.02, 0.15, 0.35); Q² predictive relevance above 0; PLSpredict for out-of-sample prediction |
| Model fit | SRMR below 0.08 (as an indication) |
Report both the measurement model and the structural model in tables, along with path diagrams.
Econometric Analysis of Secondary Data
Financial and economic studies use data from sources such as the Colombo Stock Exchange, companies' annual reports, the Central Bank of Sri Lanka, and the Department of Census and Statistics. Common methods (Wooldridge, 2016; Brooks, 2019; Gujarati & Porter, 2009):
- Panel data regression: pooled OLS, fixed effects, and random effects; the Hausman test (Hausman, 1978) helps choose between fixed and random effects; use robust standard errors for heteroscedasticity and clustering (Petersen, 2009).
- Time-series analysis: stationarity (the Augmented Dickey-Fuller test, Dickey & Fuller, 1979), cointegration, vector autoregression, Granger causality, and volatility models (GARCH).
- Event studies measure the effect of an announcement on share prices (MacKinlay, 1997).
- Diagnostics: test for multicollinearity, heteroscedasticity, autocorrelation, and endogeneity; consider robustness checks.
Analysis by Hand (Manual Methods)
Understanding the manual calculations helps you to interpret software output and to answer viva questions.
| Measure | Manual formula |
|---|---|
| Mean | x̄ = Σx ÷ n |
| Standard deviation (sample) | s = √[Σ(x − x̄)² ÷ (n − 1)] |
| Pearson correlation | r = [nΣxy − ΣxΣy] ÷ √{[nΣx² − (Σx)²][nΣy² − (Σy)²]} |
| Simple regression | b = [nΣxy − ΣxΣy] ÷ [nΣx² − (Σx)²]; a = ȳ − b x̄ |
| Independent t-test (equal variances) | t = (x̄₁ − x̄₂) ÷ [s_p √(1/n₁ + 1/n₂)] |
| Chi-square | χ² = Σ(O − E)² ÷ E; df = (rows − 1)(columns − 1) |
| Cronbach's alpha | α = [k ÷ (k − 1)] × [1 − Σs²ᵢ ÷ s²_total] |
Worked example (regression by hand). Advertising spend x: 1, 2, 3, 4, 5 and sales y: 3, 5, 6, 8, 10. Σx = 15, Σy = 32, Σxy = 113, Σx² = 55, n = 5. Then b = (5 × 113 − 15 × 32) ÷ (5 × 55 − 225) = 85 ÷ 50 = 1.7 and a = 6.4 − 1.7 × 3 = 1.3, so ŷ = 1.3 + 1.7x. This matches the SPSS result for the same data.
Worked example (chi-square by hand). In a survey of 200 customers, 60 of 100 men and 40 of 100 women prefer a brand: the expected counts are 50 and 50 in each row (total 100 preferring and 100 not). χ² = (60 − 50)²/50 + (40 − 50)²/50 + (40 − 50)²/50 + (60 − 50)²/50 = 2 + 2 + 2 + 2 = 8.0; with df = 1 the critical value at 5 percent is 3.84, so the association is significant.
Spreadsheet Tips
Use Excel for data cleaning (filters, sorting, COUNTIF), simple statistics (AVERAGE, STDEV.S, CORREL), pivot tables for cross-tabulations, and charts. Export clean data to SPSS or R. Always keep an unedited copy of the raw data.
CHAPTER 12