-
Notifications
You must be signed in to change notification settings - Fork 9
/
IV1.html
445 lines (343 loc) · 15.1 KB
/
IV1.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
<!DOCTYPE html>
<html lang="" xml:lang="">
<head>
<title>Instrumental Variable Estimation 1: Framework</title>
<meta charset="utf-8" />
<meta name="author" content="Instructor: Yuta Toyama" />
<script src="IV1_files/header-attrs-2.6/header-attrs.js"></script>
<link rel="stylesheet" href="xaringan-themer.css" type="text/css" />
</head>
<body>
<textarea id="source">
class: center, middle, inverse, title-slide
# Instrumental Variable Estimation 1: Framework
### Instructor: Yuta Toyama
### Last updated: 2021-07-02
---
class: title-slide-section, center, middle
name: logistics
# Introduction
---
## Introduction: Endogeneity Problem and its Solution
- This chapter introduces **instrumental variable (IV, 操作変数)** method as a solution to endogeneity problem.
- The lecture plan
1. More on endogeneity issues
2. Framework
3. Implementation in R
---
class: title-slide-section, center, middle
name: logistics
# Endogeneity
---
## More Examples of Endogeneity Problem
- Talk more on endogeneity problems.
1. Omitted variable bias
2. Measurement error
3. Simultaneity
---
## Example 1: More on Omitted Variable Bias
- Remember the wage regression equation (true model)
`\begin{eqnarray}
\log W_{i} &=& & \beta_{0}+\beta_{1}S_{i}+\beta_{2}A_{i}+u_{i} \\
E[u_{i}|S_{i},A_{i}] &=& & 0
\end{eqnarray}`
where `\(W_{i}\)` is wage, `\(S_{i}\)` is the years of schooling, and `\(A_{i}\)` is the ability.
- Suppose that you omit `\(A_i\)` and run the following regression instead.
`\begin{eqnarray}
\log W_{i} = alpha_{0}+\alpha_{1} S_{i} + v_i
\end{eqnarray}`
Notice that `\(v_i = \beta_2 A_i + u_i\)`, so that `\(S_i\)` and `\(v_i\)` is likely to be correlated.
---
## Does addition control variables solve OVB?
- You might want to add more and more additional variables to capture the effect of ability.
- Test scores, GPA, SAT scores, etc...
- However, how many variables should we conditioning so that `\(S_i\)` is indeed exogenous?
- Multivariate regression is valid under **selection on observables**.
- If there exists **unobserved heterogeneity** that matters both in wage and years of schooling, multivariate regression cannot deal with OVB.
---
## Example 2: Measurement error
- Reporting error, wrong understanding of the question, etc...
- Consider the regression
`$$y_{i}=\beta_{0}+\beta_{1}x_{i}^{*}+\epsilon_{i}$$`
- Here, we only observe `\(x_{i}\)` with error:
`$$x_{i}=x_{i}^{*}+e_{i}$$`
- `\(e_{i}\)` is independent from `\(\epsilon_i\)` and `\(x_i^*\)` (called classical measurement error)
- You can think `\(e_i\)` as a noise added to the data.
---
- The regression equation is
`\begin{eqnarray}
y_{i} &=& \ \beta_{0}+\beta_{1}(x_{i}-e_{i})+\epsilon_{i} \\
&=& \ \beta_{0}+\beta_{1}x_{i}+(\epsilon_{i}-\beta_{1}e_{i})
\end{eqnarray}`
- Then we have correlation between `\(x_{i}\)` and the error `\(\epsilon_{i}-\beta_{1}e_{i}\)`, violating the mean independence assumption.
- Measurement error in explantory variables leads to smaller estimates, known as **attenuation bias**.
---
## Example 3: Simultaneity (同時性) or reverse causality
- **Dependent and explanatory variable are determined simultaneously.**
- Consider the demand and supply curve
`\begin{eqnarray}
q^{d} =\beta_{0}^{d}+\beta_{1}^{d}p+\beta_{2}^{d}x+u^{d} \\
q^{s} =\beta_{0}^{s}+\beta_{1}^{s}p+\beta_{2}^{s}z+u^{s}
\end{eqnarray}`
- The equilibrium price and quantity are determined by `\(q^{d}=q^{s}\)`.
---
- In this case,
`$$p=\frac{(\beta_{2}^{s}z-\beta_{2}^{d}z)+(\beta_{0}^{s}-\beta_{0}^{d})+(u^{s}-u^{d})}{\beta_{1}^{d}-\beta_{1}^{s}}$$`
implying the correlation between the price and the error term.
- Putting this differently, the data points we observed is the intersection of these supply and demand curves.
---
- How can we distinguish demand and supply?
![Demand Supply](figs/fig_Demand_Supply.png)
---
## Causal Effect of Price on Quantity?
- What is a causal effect of price on quantity? What is the sign (if it exists)?
- For consumer: Higher prices leads to lower demand.
- For (price-taking) firms: Higher prices leads to more supply.
- "Causal effect of price on quantity" is a meaningless concept unless you specify either demand or supply.
---
class: title-slide-section, center, middle
name: logistics
# IV Idea
---
## Idea of IV Regression
- Let's start with a simple case.
`$$y_i = \beta_0 + \beta_1 x_i + \epsilon_i, Cov(x_i, \epsilon_i) \neq 0$$`
- Define the variable `\(z_i\)` named **instrumental variable (IV)** that satisfies the following conditions:
1. **Independence (独立性)**: `\(Cov(z_i, \epsilon_i) = 0\)`. No correlation between IV and error.
2. **Relevance (関連性)**: `\(Cov(z_i, x_i) \neq 0\)`. Correlation between IV and endogenous variable `\(x_i\)`.
- Idea: Use the variation of `\(x_i\)` **induced by instrument `\(z_i\)`** to estimate the direct (causal) effect of `\(x_i\)` on `\(y_i\)`, that is `\(\beta_1\)`!.
---
## More on Idea
1. Intuitively, the OLS estimator captures the correlation between `\(x\)` and `\(y\)`.
2. If there is no correlation between `\(x\)` and `\(\epsilon\)`, it captures the causal effect `\(\beta_1\)`.
3. If not, the OLS estimator captures both direct and indirect effect, the latter of which is bias.
4. Now, let's capture the variation of `\(x\)` due to instrument `\(z\)`,
- Such a variation should exist under **relevance** assumption.
- Such a variation should not be correlated with the error under **independence assumption**
5. By looking at the correlation between such variation and `\(y\)`, you can get the causal effect `\(\beta_1\)`.
---
.middle[
.center[
<img src="figs/fig_IV_idea.png" width="650">
]
]
---
## Identification of Parameter with IV
- Taking the covariance between `\(y_i\)` and `\(z_i\)`
$$
Cov(y_i, z_i) = \beta_1 Cov(x_i, z_i) + Cov(\epsilon_i, z_i )
$$
- IV conditions implie that
$$
\beta_1 = \frac{Cov(y_i, z_i)}{Cov(x_i, z_i)}
$$
- Question: Can you see the roles of IV conditions?
---
class: title-slide-section, center, middle
name: logistics
# IV General Framework
---
## Multiple endogenous variables and instruments
- Consider the model
`\begin{eqnarray}
Y_i = \beta_0 + \beta_1 X_{1i} + \dots + \beta_K X_{Ki} + \beta_{K+1} W_{1i} + \dots + \beta_{K+R} W_{Ri} + u_i,
\end{eqnarray}`
- `\(Y_i\)` is the dependent variable
- `\(X_{1i},\dots,X_{Ki}\)` are `\(K\)` endogenous (内生) regressors: `\(Cov(X_{ki}, u_i) \neq 0\)` for all `\(k\)`.
- `\(W_{1i},\dots,W_{Ri}\)` are `\(R\)` exogenous (外生) regressors which are uncorrelated with `\(u_i\)`. `\(Cov(W_{ri}, u_i) = 0\)` for all `\(r\)`.
- `\(u_i\)` is the error term
- `\(Z_{1i},\dots,Z_{Mi}\)` are `\(M\)` instrumental variables
- `\(\beta_0,\dots,\beta_{K+R}\)` are `\(1+K+R\)` unknown regression coefficients
---
## Two Stage Least Squares (2SLS, 二段階最小二乗法)
- Step 1: **First-stage regression(s)**
- For each of the endogenous variables ( `\(X_{1i},\dots,X_{ki}\)` ), run an OLS regression on all IVs ( `\(Z_{1i},\dots,Z_{mi}\)` ), all exogenous variables ( `\(W_{1i},\dots,W_{ri}\)` ) and an intercept.
- Compute the fitted values ( `\(\widehat{X}_{1i},\dots,\widehat{X}_{ki}\)` ).
- Step 2: **Second-stage regression**
- Regress the dependent variable `\(Y_i\)` on **the predicted values** of all endogenous regressors ( `\(\widehat{X}_{1i},\dots,\widehat{X}_{ki}\)` ), all exogenous variables ( `\(W_{1i},\dots,W_{ri}\)`) and an intercept using OLS.
- This gives `\(\widehat{\beta}_{0}^{TSLS},\dots,\widehat{\beta}_{k+r}^{TSLS}\)`, the 2SLS estimates of the model coefficients.
---
## Why 2SLS works
- Consider a simple case with 1 endogenous variable and 1 IV.
- In the first stage, we estimate
`$$x_i = \pi_0 + \pi_1 z_i + v_i$$`
by OLS and obtain the fitted value `\(\widehat{x}_i = \widehat{\pi}_0 + \widehat{\pi}_1 z_i\)`.
- In the second stage, we estimate
`$$y_i = \beta_0 + \beta_1 \widehat{x}_i + u_i$$`
- Since `\(\widehat{x}_i\)` depends only on `\(z_i\)`, which is uncorrelated with `\(u_i\)`, the second stage can estimate `\(\beta_1\)` without bias.
---
class: title-slide-section, center, middle
name: logistics
# Conditions for IV
---
## Conditions for Valid IVs: Necessary condition
- Depending on the number of IVs, we have three cases
1. Over-identification (過剰識別): `\(M > K\)`
2. Just identification (丁度識別): `\(M=K\)`
3. Under-identification (過小識別): `\(M < K\)`
- The necessary condition is `\(M \geq K\)`.
- We should have more IVs than endogenous variables.
---
## Sufficient condition
- In a general framework, the sufficient conditions for valid instruments
1. **Independence**: `\(Cov( Z_{mi}, u_i) = 0\)` for all `\(m\)`.
2. **Relevance**: In the second stage regression, the variables
`$$\left( \widehat{X}_{1i},\dots,\widehat{X}_{ki}, W_{1i},\dots,W_{ri}, 1 \right)$$`
are **not perfectly multicollinear**.
- What does the relevance condition mean?
---
## Relevance condition
- In the simple example above, The first stage is
`$$x_i = \pi_0 + \pi_1 z_i + v_i$$`
and the second stage is
`$$y_i = \beta_0 + \beta_1 \widehat{x}_i + u_i$$`
- The second stage would have perfect multicollinarity if `\(\pi_1 = 0\)` (i.e., `\(\widehat{x}_i = \pi_0\)`).
---
<br/><br/>
- Back to the general case, the first stage for `\(X_k\)` can be written as
`$$X_{ki} = \pi_0 + \pi_1 Z_{1i} + \cdots + \pi_M Z_{Mi} + \pi_{M+1} W_{1i} + \cdots + \pi_{M+R} W_{Ri}$$`
and one of `\(\pi_1 , \cdots, \pi_M\)` should be non-zero.
- Intuitively speaking, **the instruments should be correlated with endogenous variables after controlling for exogenous variables**
---
## Check Instrument Validity: Relevance
- **Weak instruments (弱操作変数)** if those instruments explain little variation in the endogenous variables.
- Weak instruments lead to
1. imprecise estimates (higher standard errors)
2. The asymptotic distribution would deviate from a normal distribution even if we have a large sample.
---
## A rule of thumb to check the relevance conditions.
- Consider the case with one endogenous variable `\(X_{1i}\)`.
- The first stage regression
`$$X_k = \pi_0 + \pi_1 Z_{1i} + \cdots + \pi_M Z_{Mi} + \pi_{M+1} W_{1i} + \cdots + \pi_{M+R} W_{Ri}$$`
- Conduct F test.
`\begin{eqnarray}
H_0 & : \pi_1 = \cdots = \pi_M = 0 \\
H_1 & : otherwise
\end{eqnarray}`
- If we can reject this, we can say no concern for weak instruments.
- A rule of thumbs is that the F statistic should be larger than 10. (Stock, Wright, and Yogo 2012)
---
## Independence (Instrument exogeneity)
- This is essentially non-testable assumption, as in mean independence assumption in OLS.
- Justification of this assumption depends on a context, institutional features, etc...
</textarea>
<style data-target="print-only">@media screen {.remark-slide-container{display:block;}.remark-slide-scaler{box-shadow:none;}}</style>
<script src="https://remarkjs.com/downloads/remark-latest.min.js"></script>
<script>var slideshow = remark.create({
"highlightStyle": "github",
"highlightLines": true,
"countIncrementalSlides": false,
"ratio": "16:9"
});
if (window.HTMLWidgets) slideshow.on('afterShowSlide', function (slide) {
window.dispatchEvent(new Event('resize'));
});
(function(d) {
var s = d.createElement("style"), r = d.querySelector(".remark-slide-scaler");
if (!r) return;
s.type = "text/css"; s.innerHTML = "@page {size: " + r.style.width + " " + r.style.height +"; }";
d.head.appendChild(s);
})(document);
(function(d) {
var el = d.getElementsByClassName("remark-slides-area");
if (!el) return;
var slide, slides = slideshow.getSlides(), els = el[0].children;
for (var i = 1; i < slides.length; i++) {
slide = slides[i];
if (slide.properties.continued === "true" || slide.properties.count === "false") {
els[i - 1].className += ' has-continuation';
}
}
var s = d.createElement("style");
s.type = "text/css"; s.innerHTML = "@media print { .has-continuation { display: none; } }";
d.head.appendChild(s);
})(document);
// delete the temporary CSS (for displaying all slides initially) when the user
// starts to view slides
(function() {
var deleted = false;
slideshow.on('beforeShowSlide', function(slide) {
if (deleted) return;
var sheets = document.styleSheets, node;
for (var i = 0; i < sheets.length; i++) {
node = sheets[i].ownerNode;
if (node.dataset["target"] !== "print-only") continue;
node.parentNode.removeChild(node);
}
deleted = true;
});
})();
(function() {
"use strict"
// Replace <script> tags in slides area to make them executable
var scripts = document.querySelectorAll(
'.remark-slides-area .remark-slide-container script'
);
if (!scripts.length) return;
for (var i = 0; i < scripts.length; i++) {
var s = document.createElement('script');
var code = document.createTextNode(scripts[i].textContent);
s.appendChild(code);
var scriptAttrs = scripts[i].attributes;
for (var j = 0; j < scriptAttrs.length; j++) {
s.setAttribute(scriptAttrs[j].name, scriptAttrs[j].value);
}
scripts[i].parentElement.replaceChild(s, scripts[i]);
}
})();
(function() {
var links = document.getElementsByTagName('a');
for (var i = 0; i < links.length; i++) {
if (/^(https?:)?\/\//.test(links[i].getAttribute('href'))) {
links[i].target = '_blank';
}
}
})();
// adds .remark-code-has-line-highlighted class to <pre> parent elements
// of code chunks containing highlighted lines with class .remark-code-line-highlighted
(function(d) {
const hlines = d.querySelectorAll('.remark-code-line-highlighted');
const preParents = [];
const findPreParent = function(line, p = 0) {
if (p > 1) return null; // traverse up no further than grandparent
const el = line.parentElement;
return el.tagName === "PRE" ? el : findPreParent(el, ++p);
};
for (let line of hlines) {
let pre = findPreParent(line);
if (pre && !preParents.includes(pre)) preParents.push(pre);
}
preParents.forEach(p => p.classList.add("remark-code-has-line-highlighted"));
})(document);</script>
<script>
slideshow._releaseMath = function(el) {
var i, text, code, codes = el.getElementsByTagName('code');
for (i = 0; i < codes.length;) {
code = codes[i];
if (code.parentNode.tagName !== 'PRE' && code.childElementCount === 0) {
text = code.textContent;
if (/^\\\((.|\s)+\\\)$/.test(text) || /^\\\[(.|\s)+\\\]$/.test(text) ||
/^\$\$(.|\s)+\$\$$/.test(text) ||
/^\\begin\{([^}]+)\}(.|\s)+\\end\{[^}]+\}$/.test(text)) {
code.outerHTML = code.innerHTML; // remove <code></code>
continue;
}
}
i++;
}
};
slideshow._releaseMath(document);
</script>
<!-- dynamically load mathjax for compatibility with self-contained -->
<script>
(function () {
var script = document.createElement('script');
script.type = 'text/javascript';
script.src = 'https://mathjax.rstudio.com/latest/MathJax.js?config=TeX-MML-AM_CHTML';
if (location.protocol !== 'file:' && /^https?:/.test(script.src))
script.src = script.src.replace(/^https?:/, '');
document.getElementsByTagName('head')[0].appendChild(script);
})();
</script>
</body>
</html>