social-science-analysis

# Social Science Analysis

Research methods for social and behavioral sciences. Venv: `source /Users/zhangmingda/clawd/.venv/bin/activate`

## Survey Design

### Question Types & Best Practices
- **Closed-ended**: Likert scales, multiple choice, ranking
- **Open-ended**: Free text (harder to analyze, richer data)
- **Matrix questions**: Multiple items, same scale (efficient but watch for straight-lining)

### Likert Scale Design
```
Strongly Disagree (1) — Disagree (2) — Neutral (3) — Agree (4) — Strongly Agree (5)
```
- Use 5 or 7 points (odd for neutral option)
- Mix positively and negatively worded items (reverse-code in analysis)
- Avoid double-barreled questions
- Pilot test with 10-20 respondents

### Sampling Methods
| Method | When | Pros | Cons |
|--------|------|------|------|
| Simple random | Known population | Unbiased | Need sampling frame |
| Stratified | Subgroup comparison | Precise estimates per stratum | Complex |
| Cluster | Geographic spread | Cost-effective | Higher design effect |
| Convenience | Exploratory | Easy | Not generalizable |
| Snowball | Hard-to-reach populations | Access hidden groups | Selection bias |
| Quota | Ensure representation | Practical | Not truly random |

## Psychometrics & Scale Development

### Reliability
```python
import numpy as np

def cronbachs_alpha(items_df):
    """Calculate Cronbach's alpha for scale reliability"""
    k = items_df.shape[1]
    item_vars = items_df.var(axis=0, ddof=1)
    total_var = items_df.sum(axis=1).var(ddof=1)
    alpha = (k / (k - 1)) * (1 - item_vars.sum() / total_var)
    return alpha

# Interpretation: α > 0.7 acceptable, > 0.8 good, > 0.9 excellent
```

### Exploratory Factor Analysis
```python
from sklearn.decomposition import FactorAnalysis
import numpy as np

# Determine number of factors (parallel analysis or scree plot)
fa = FactorAnalysis(n_components=3, rotation='varimax')
fa.fit(X_scaled)
loadings = pd.DataFrame(fa.components_.T, index=item_names, columns=['F1', 'F2', 'F3'])
print(loadings.round(3))
# Items loading > 0.4 on a factor belong to that construct
```

### Confirmatory Factor Analysis / SEM
For CFA and SEM, recommend using R with `lavaan` package or Python `semopy`:
```python
# pip install semopy
import semopy

model_spec = """
    Latent1 =~ item1 + item2 + item3
    Latent2 =~ item4 + item5 + item6
    Latent1 ~ Latent2
"""
model = semopy.Model(model_spec)
model.fit(df)
print(model.inspect())
# Check fit indices: CFI > 0.95, RMSEA < 0.06, SRMR < 0.08
```

## Qualitative Analysis

### Thematic Analysis (Braun & Clarke)
1. **Familiarization**: Read and re-read data
2. **Initial coding**: Generate codes systematically
3. **Theme search**: Collate codes into potential themes
4. **Theme review**: Check themes against coded extracts and full dataset
5. **Theme definition**: Name and define each theme
6. **Report**: Select vivid examples, relate to research question

### Coding Framework Template
```markdown
| Code | Definition | Example Quote | Theme |
|------|-----------|---------------|-------|
| ADAPT | Adaptation strategy | "We had to change our approach..." | Resilience |
| BARR | Barrier encountered | "The main obstacle was..." | Challenges |
```

### Grounded Theory
1. Open coding → Axial coding → Selective coding
2. Constant comparison method
3. Theoretical sampling until saturation
4. Memo writing throughout

## Content Analysis

```python
# Quantitative content analysis
import pandas as pd
from collections import Counter

def content_analysis(texts, codebook):
    """
    codebook: dict of {category: [keywords]}
    Returns frequency matrix
    """
    results = []
    for text in texts:
        text_lower = text.lower()
        counts = {}
        for category, keywords in codebook.items():
            counts[category] = sum(text_lower.count(kw.lower()) for kw in keywords)
        results.append(counts)
    return pd.DataFrame(results)

# Inter-coder reliability (Cohen's Kappa)
from sklearn.metrics import cohen_kappa_score
kappa = cohen_kappa_score(coder1_labels, coder2_labels)
# κ > 0.8 excellent, 0.6-0.8 substantial, 0.4-0.6 moderate
```

## Social Network Analysis

```python
import networkx as nx
import numpy as np

G = nx.from_pandas_edgelist(df, 'source', 'target')

# Centrality measures
degree = nx.degree_centrality(G)
betweenness = nx.betweenness_centrality(G)
closeness = nx.closeness_centrality(G)
eigenvector = nx.eigenvector_centrality(G)

# Community detection
from networkx.algorithms.community import greedy_modularity_communities
communities = list(greedy_modularity_communities(G))

# Network statistics
print(f"Nodes: {G.number_of_nodes()}, Edges: {G.number_of_edges()}")
print(f"Density: {nx.density(G):.4f}")
print(f"Clustering coefficient: {nx.average_clustering(G):.4f}")
```

## Tips
- Pre-register hypotheses and analysis plans (OSF, AsPredicted)
- Report Cronbach's alpha for all scales
- Use power analysis for sample size determination
- For qualitative research, document your positionality
- Mixed methods: clearly state the integration strategy
- IRB/ethics approval is mandatory for human subjects research