.hi-green[Electoral Systems and Geographically-Targeted Oversight:]

class: center, middle, inverse, title-slide

.title[
# .hi-green[Electoral Systems and Geographically-Targeted Oversight:]
]
.subtitle[
## .smallest.hi-slate[Evidence from Taiwan Legislative Yuan]
]
.author[
### <strong>Yen-Chieh Liao and Li Tang</strong>
]
.author[
### .tiny[University of Birmingham and University of Reading]
]
.author[
### .tiny[2025 APSA]
]
.date[
### .tiny[13 September 2025]
]

---

exclude: true

---
layout: true
# .tiny[Overview]
---
name:overview

- .small.hi-grey[**Motivation & Research Question:**]<br>

.small[How does Taiwan's SNTV→SMD electoral reform affect legislators' particularistic behaviors?]
  
--

- .small.hi-grey[**Data:**] <br>
  + .small[63,748 parliamentary questions from 402 district legislators (two decades)]<br>

--
  + .small[Focus on geographically targeted questions (GTQs) — .blue[__content that addresses constituency-specific geographical interests and demands__]]
  
--

- .small.hi-grey[**Method & Empirical Strategy:**] <br> 
  .small[Fine-tuned transformer models to identify geographically targeted content + regression analysis]

--
  
- .small.hi-grey[**Major Finding:**] <br> 
  + .small[SNTV → SMD transition reduces GTQs, with variation across municipalities by socioeconomic characteristics] <br> 
  + .small[SNTV demonstrates greater particularistic responsiveness than SMD]
  
---
layout: true
# .tiny[Motivation]
---
name:why-pq

__What We Know:__

- .small.hi-grey[Established:] <br>
Different candidate-centered systems (SNTV v.s. SMD) create varying incentives for personal reputation (e.g., Shugart and Wattenberg, 2001; Shugart, 2005; Grofman et al., 1999).<br>

- .small.hi-grey[Electoral System Literature:] 
Studies on manifestos (e.g., Catalinac, 2016a,b; Crisp et al., 2024), legislative voting (e.g., Liao, 2024), political deliberation (e.g., Motolinia, 2021; Høyland and Søyland, 2019) and legislation (e.g., Decadri et al., 2022).

__Gap__

- .small.hi-grey[**Missing:**] Previous studies focus on legislative outputs/campaign promises, not ongoing behavior

- .small.hi-grey[**Opportunity:**] Systematic evidence on SNTV→SMD transitions Japan <br>  
--
  → Taiwan's reform provides anotehr ideal case study

---
layout: true
---
name:why-pq

### .center[*Two important debates in candidate-centered electoral system literature*]

---
layout: true
# .tiny[Debates Ⅰ]
---
name:why-pq

__Does SNTV → SMD Reduce Particularistm?__

.small.hi-grey[**The Literature:**] 
  - SNTV's intraparty competition encourages  →  personal reputation  →  particularistic appeals<br>
--

- SMD's interparty competition promotes programmatic appeals to broader coalition .small[(e.g., Cain 1987; Carey 1995; Cox 1997; Duverger 1954; Catalinac 2016)]

.small.hi-grey[**Japan Evidence:**] Shift from pork-barrel to programmatic campaigning .small[(Catalinac 2016) and issue coverge on copartisan on manifesto (Catalinac 2017)]

.small.hi-grey[**Taiwan Evidence:**] Particularistic behavior persists or increases .small[(Liao et al. 2013; Luor and Shieh 2009)]

> **Hypothesis 1.** Due to the elimination of intraparty competition and the necessity to appeal to broader constituencies, __legislators elected under MMM are less likely to submit GTQs compared to those elected under the SNTV system.__

---
layout: true
# .tiny[Dabate ⅠⅠ]
---
name:why-pq

&nbsp;

__Do Geographic Conditions Moderate Electoral System Effects?__

- .small.hi-grey[Local Socioeconomic Conditions:]<br> 
Economic challenges/homogeneity may override institutional incentives

- .small.hi-grey[Evidence:] <br>
Unemployment, manufacturing concentration affect legislative behavior .small[(Rickard 2012; Cayton 2022; Harden 2013)]

&nbsp;&nbsp;
&nbsp;&nbsp;

> **Hypothesis 2.** While MMM reduces GTQs compared to SNTV, __this effect is moderated by local socioeconomic characteristics__.

---
layout: true
# .tiny[Data: Two Decades of Parliamentary Questions]
---
name:why-pq

__Our Dataset__
- .small.hi-grey[**Scope:**] 63,748 parliamentary questions from 402 district legislators (20-year period) from SNTV to MMM, focusing on district constituencies only (excluding CLPR and indigenous constituencies)

__Why Parliamentary Questions?__
- .small.hi-grey[**Advantages:**] Continuous evidence of policy priorities throughout terms .small[(Martin 2011; Saalfeld 2011)]
- .small.hi-grey[**vs. Floor Debates:**] No time limits or strict party discipline constraints .small[(Proksch 2011; Martin 2010)]
- .small.hi-grey[**vs. Manifestos:**] Captures actual legislative behavior, not just campaign promises .small[(Stafford 2025)]

__Taiwan's Institutional Design__
- .small.hi-grey[**Compusalsoiry:**] Mandatory 20-day response requirement enhances strategic use
- .small.hi-grey[**Build personal reputation:**]  national policy vs local constituency needs

---
layout: true
# .tiny[Transformer-Based Detection of GTQ Content]
---
name:why-pq

.pull-left[
<img src="./images/train-data.png" width="95%" style="display: block; margin: auto;" />
]

.pull-right[

<img src="./images/train-split.png" width="100%" style="display: block; margin: auto;" />
]

---

---

---

---

---

.pull-left[
<img src="./images/train-data.png" width="95%" style="display: block; margin: auto;" />
]

.pull-right[

<img src="./images/train-split.png" width="100%" style="display: block; margin: auto;" />
]

.pull-left[

<img src="./images/performance.png" width="99%" style="display: block; margin: auto;" />
]

---

<br>

---

<br>

---

<br>

---

<br>

---

<br>

---

.pull-left[
<img src="./images/train-data.png" width="95%" style="display: block; margin: auto;" />
]

.pull-right[

<img src="./images/train-split.png" width="100%" style="display: block; margin: auto;" />
]

.pull-left[

<img src="./images/performance.png" width="99%" style="display: block; margin: auto;" />
]

--
.pull-right[
<img src="./images/PQ.png" width="90%" style="display: block; margin: auto;" />
]

---
layout: true
# .tiny[Performance on Held-out data]
---
name:why-pq

.pull-left[
<img src="./images/val-set.png" width="95%" style="display: block; margin: auto;" />
]

.pull-right[
<img src="./images/val-setb.png" width="95%" style="display: block; margin: auto;" />
]

---

.pull-left[
<img src="./images/val-set2.png" width="95%" style="display: block; margin: auto;" />
]

.pull-right[
<img src="./images/val-setb2.png" width="95%" style="display: block; margin: auto;" />
]

---
layout: true
# .tiny[Posthoc Validation]
---
name:data

---

---

---

---
layout: true
# .tiny[Dependent Variable: Proportion of GTQs]
---
name:dv
<br><br>

.pull-left[
.small[**Definition**]
`$$\small{\text{Proportion of GTQs}_{i,t} = \frac{\text{GTQs}_{i,t}}{\text{Total Questions}_{i,t}} \times 100}$$`

.small[**Refine**]
- .small[63,748 total parliamentary questions]
- .small[8,992 high-confidence GTQs (14%)]
- .small[95% confidence threshold for precision]
]

.pull-right[
.small[**Sample Coverage**]
- .small[**Unit of Analysis**: Legislator-year]
- .small[**Time Period**: 1999-2019]
  - .small[SNTV period: 1999-2008]
  - .small[MMM period: 2008-2019]
- .small[**Final Sample**: 2,809 district observations]
- .small[Excludes CLPR and indigenous constituencies]
]

---

.pull-left[
.small[**Definition**]
`$$\small{\text{Proportion of GTQs}_{i,t} = \frac{\text{GTQs}_{i,t}}{\text{Total Questions}_{i,t}} \times 100}$$`

.small[**Refine**]
- .small[63,748 total parliamentary questions]
- .small[8,992 high-confidence GTQs (14%)]
- .small[95% confidence threshold for precision]
]

.pull-right[
.small[**Sample Coverage**]
- .small[**Unit of Analysis**: .blue[__Legislator-year__]]
- .small[**Time Period**: 1999-2019]
  - .small[SNTV period: 1999-2008]
  - .small[MMM period: 2008-2019]
- .small[**Final Sample**: 2,809 district observations]
- .small[Excludes CLPR and indigenous constituencies]
]

---
layout: true
# .tiny[Regression Design]
---
name:research-design

.pull-left[
**Part 1: Electoral System Effects**

- `$$\small{\text{GTQs}_{i,t} = \alpha_{0} + \alpha_{1}\text{Post-reform}_{t} +\\ \alpha_{2}\text{Year} + \alpha_{3}(\text{Post-reform}_{t} \times \text{Year}) + \\ \theta\text{Controls}_{i,t} + \mu_{i} + \epsilon_{i,t}}$$`
- .small[**Municipality FE +  Controls**: Party, seniority, gender, pork committee membership, electoral margin, district magnitude]
]

.pull-right[
**Part 2: Socioeconomic Moderation**

- `$$\small{\text{GTQs}_{i,t} = \alpha_{0} + \alpha_{1}\text{Post-reform}_{t} + \\ \sum_{d=1}^{5}\alpha_{4}^d(\text{Post-reform}_{t} \times \text{Conditions}^{d}_{t}) +\\ \text{Controls}}$$`

- .small[**5 Municipal Indicators**:]
  - .small[Unemployment rate]
  - .small[Sector employment (Agri/Manu/Service)]
  - .small[Higher education population share]
]

---

.pull-left[
**Part 1: Electoral System Effects**

- `$$\small{\text{GTQs}_{i,t} = \alpha_{0} + \alpha_{1}\text{Post-reform}_{t} +\\ \alpha_{2}\text{Year} + \alpha_{3}(\text{Post-reform}_{t} \times \text{Year}) + \\ \theta\text{Controls}_{i,t} + \mu_{i} + \epsilon_{i,t}}$$`
- .small[**Municipality FE +  Controls**: .blue[__Party, seniority, gender, pork committee membership, electoral margin, district magnitude__]]
]

.pull-right[
**Part 2: Socioeconomic Moderation**

- `$$\small{\text{GTQs}_{i,t} = \alpha_{0} + \alpha_{1}\text{Post-reform}_{t} + \\ \sum_{d=1}^{5}\alpha_{4}^d(\text{Post-reform}_{t} \times \text{Conditions}^{d}_{t}) +\\ \text{Controls}}$$`

- .small[**5 Municipal Indicators**:]
  - .small[Unemployment rate, Agricultural value]
  - .small[Sector employment (Agri/Manu/Service)]
  - .small[Higher education population share]
]

---

.pull-left[
**Part 1: Electoral System Effects**

- `$$\small{\text{GTQs}_{i,t} = \alpha_{0} + \alpha_{1}\text{Post-reform}_{t} +\\ \alpha_{2}\text{Year} + \alpha_{3}(\text{Post-reform}_{t} \times \text{Year}) + \\ \theta\text{Controls}_{i,t} + \mu_{i} + \epsilon_{i,t}}$$`
- .small[**Municipality FE +  Controls**: .blue[__Party, seniority, gender, pork committee membership, electoral margin, district magnitude__]]
]

.pull-right[
**Part 2: Socioeconomic Moderation**

- `$$\small{\text{GTQs}_{i,t} = \alpha_{0} + \alpha_{1}\text{Post-reform}_{t} + \\ \sum_{d=1}^{5}\alpha_{4}^d(\text{Post-reform}_{t} \times \text{Conditions}^{d}_{t}) +\\ \text{Controls}}$$`

- .small[**5 Municipal Indicators**:]
  - .small[.blue[__Unemployment rate, Agricultural value__]]
  - .small[.blue[__Sector employment (Agri/Manu/Service)__]]
  - .small[.blue[__Higher education population share__]]
]

---
layout: true
# .tiny[Main Findings: The Effects on Particularistic Behavior (1/2)]
---
<div style="margin-top: -20px;">

.pull-left[
<div style="margin-top: 80px;">

]

.pull-right[
**SNTV (Pre-2008)**
- .small[Legislators differentiate via distributive activities]
- .small[Higher GTQs]

**SMD  (Post-2008)**
- .small[SMD → Interparty competition]
- .small[Lower GTQ initially]
- .small[Party labels ↑]
- .small[Effect diminishes over time]
]

---

.pull-left[
<div style="margin-top: 80px;">

]

.pull-right[
**SNTV (Pre-2008)**
- .small[Legislators differentiate via distributive activities]
- .small[Higher GTQs]

**SMD  (Post-2008)**
- .small[Lower GTQ initially]
- .small[SMD → Interparty competition]
- .small[Party labels ↑]
- .small.blue[__Effect diminishes over time__]
]

---
layout: true
# .tiny[Main Findings: The Impacts of Socioeconomic Context (2/2)]
---
<div style="margin-top: -20px;">

.pull-left[
<div style="margin-top: 80px;">
<img src="./images/result2.png" width="95%" style="display: block; margin: auto;" />
</div>
]

.pull-right[
**Under SNTV**
- .small[High unemployment: Fewer distributive policies]
- .small[Economic stability: More GTQs for reputation]
- .small[Lower education: Traditional patron-client relations]
- .small[Higher education: Less distributive engagement]

**Under SMD**
- .small[Both socioeconomic effects substantially neutralized]
- .small[Single-member structure reduces contextual responsiveness]
]

---

.pull-left[
<div style="margin-top: 80px;">
<img src="./images/result2.png" width="95%" style="display: block; margin: auto;" />
</div>
]

.pull-right[
**Under SNTV**
- .small[.blue[**High unemployment**]: Fewer distributive policies]
- .small[.blue[**Economic stability**]: More GTQs for reputation]
- .small[.blue[**Lower education**]:  Traditional patron-client relations]
- .small[.blue[**Higher education**]: Less distributive engagement]

**Under SMD**
- .small[Both socioeconomic effects substantially neutralized]
- .small[Single-member structure reduces contextual responsiveness]
]

---

.pull-left[
<div style="margin-top: 80px;">
<img src="./images/result2.png" width="95%" style="display: block; margin: auto;" />
</div>
]

**Under SMD**
- .small[Both socioeconomic effects substantially .blue[**neutralized**]]
- .small[Single-member structure .blue[**reduces**] contextual responsiveness]
]

---
layout: true
# .tiny[Takeaway]
---

**Reform Goals (KMT + DPP): SNTV → SMD**<br>
- .small[Reduce money politics and factionalism]<br> 
- .small[Improve legislative efficiency]

**Actual Effects: SMD Impact**<br> 
- .small[Socioeconomic effects substantially neutralized]<br> 
- .small[SMD reduces responsiveness]

**Academic Debate → Reality**<br> 
- .small[Wu (2002): SNTV was unfairly scapegoated for broader political problems]<br> 
- .small[We: SMD reduces particularistic responsiveness]<br>

---
layout: true
class: inverse, center, middle
# Thank You
---