Strength In Numbers: How Polls Work and Why We Need Them

class: left, top, title-slide

# Strength In Numbers: How Polls Work and Why We Need Them
## A first look at my upcoming book (which is really just as much about democracy as it is about polls)
### G. Elliott Morris | December 2, 2021

---

---

---

---

---

---

.center[
# Surveys as tools
]

### ... for what? sociology?

## democracy?

Can we use polls to measure what people actually **want**, and then have the government do whatever that is

(... conditional on it being reasonable, thought-out, and “good” — which we will get to later)

---

# Strength in numbers

1. Arithmetic
2. Democratic

(Actually we're gonna tackle these in reverse order)

---

# Public opinion and the state

---

---

---

## “A man may not be able to make a poem, but he can tell when a poem pleases him. He may not be able to make a house, but he can tell when the roof leaks. He may not be able to cook, but he can tell whether he likes what is prepared for him. “
### — Aristotle

---

---

---

---

## “Public opinion sets bounds to every government, and is the real sovereign in every free one.”
### — James Madison

## “Where the law of the majority ceases to be acknowledged, there government ends, the law of the strongest takes its place.” 
### — Thomas Jefferson

---

.center[
&nbsp;

&nbsp;

# Polls as a rolling referendum
]

---

---

---

.center[
&nbsp;

&nbsp;

# The birth of the poll
]

---

---

---

---

# Polls today

## The same as they ever were?

---

Not so fast...

---

---

---

---

---

---

# Why polls off in 1948?

### 1. Late change
### 2. Selection bias from interviewers?
### 3. Missing weight? Education, etc
#### There had been a residual pro-R bias for years...

# Solution?

SSRC says.... **Random sampling**

---

---

---

# After 1952...

### (Gallup's) Polls used quotas

### But interviews were (more) randomized

### And attempted a partisan balance by geogprahy

---

# 1970: "Design-based" polls

---

---

---

---

# What do we do when we have doubts about the accuracy of an entire mode of data collection?

---

---

# What tweaks to the survey process could we make to increase the representativeness of respondents, conditional on having low response rates?

---

---

---

# Mister P

---

# Polls in 2016 and 2020

---

# 2016: Education weighting

---

# 2020: Partisan non-response

---
# The future?

## - CNN, Pew and others shifting to "address-based" sampling

## - Success of mixed modes

## - But a lot more work to be done...

---
# Bias in election forecasting

---

## If polls are biased, the models will be too

---

---

# The next 90 years

## Think less about prediction and more about communication
- Polling is an uncertain science
- Aggregation and forecasting hasn’t achieved the level of communication we wanted
- So we must emphasize uncertainty
- And the normative value of polls
- So that we can continue to use them to give a voice to the peopel
- Which is why they were really popularized in the first place

---

## One more thing...

---

---

# Thank you!

#### Website: [gelliottmorris.com](https://www.gelliottmorris.com)

#### Twitter: [@gelliottmorris](http://www.twitter.com/gelliottmorris)

<hr>

_These slides were made using the `xaringan` package for R. They are available online at https://www.gelliottmorris.com/slides/_