Professional Documents
Culture Documents
Chapter 2 and 3
Sebastiano Manzan
ECON-UB251
1 / 70
Brief Overview of the Course
2 / 70
This course is about using data to measure causal effects
3 / 70
In this course you will
4 / 70
Review of Probability and Statistics (SW Chapters 2, 3)
▶ Policy question: What is the effect on test scores (or some other
outcome measure) of reducing class size by one student per class? by
8 students/class?
▶ We must use data to find out (is there any way to answer this without
data?)
5 / 70
The California Test Score Data Set
▶ This table does not tell us anything about the relationship between
test scores and the STR
6 / 70
Do districts with smaller classes have higher test scores?
7 / 70
R
library(readxl)
caschool <- read_xlsx("./data/caschool.xlsx")
dplyr::glimpse(caschool)
Rows: 420
Columns: 18
$ `Observation Number` <dbl> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15~
$ dist_cod <dbl> 75119, 61499, 61549, 61457, 61523, 62042, 68536, ~
$ county <chr> "Alameda", "Butte", "Butte", "Butte", "Butte", "F~
$ district <chr> "Sunol Glen Unified", "Manzanita Elementary", "Th~
$ gr_span <chr> "KK-08", "KK-08", "KK-08", "KK-08", "KK-08", "KK-~
$ enrl_tot <dbl> 195, 240, 1550, 243, 1335, 137, 195, 888, 379, 22~
$ teachers <dbl> 10.90, 11.15, 82.90, 14.00, 71.50, 6.40, 10.00, 4~
$ calw_pct <dbl> 0.5102, 15.4167, 55.0323, 36.4754, 33.1086, 12.31~
$ meal_pct <dbl> 2.0408, 47.9167, 76.3226, 77.0492, 78.4270, 86.95~
$ computer <dbl> 67, 101, 169, 85, 171, 25, 28, 66, 35, 0, 86, 56,~
$ testscr <dbl> 690.80, 661.20, 643.60, 647.70, 640.85, 605.55, 6~
$ comp_stu <dbl> 0.34358975, 0.42083332, 0.10903226, 0.34979424, 0~
$ expn_stu <dbl> 6384.911, 5099.381, 5501.955, 7101.831, 5235.988,~
$ str <dbl> 17.88991, 21.52466, 18.69723, 17.35714, 18.67133,~
$ avginc <dbl> 22.690001, 9.824000, 8.978000, 8.978000, 9.080333~
$ el_pct <dbl> 0.000000, 4.583333, 30.000002, 0.000000, 13.85767~
$ read_scr <dbl> 691.6, 660.5, 636.3, 651.9, 641.8, 605.7, 604.5, ~
$ math_scr <dbl> 690.0, 661.9, 650.9, 643.5, 639.9, 605.4, 609.0, ~
8 / 70
mean(caschool$testscr)
[1] 654.1565
sd(caschool$testscr)
[1] 19.05335
c(mean(caschool$str), sd(caschool$str))
9 / 70
library(ggplot2)
ggplot(caschool, aes(str, testscr)) + geom_point()
testscr 690
660
630
14 16 18 20 22 24 26
str
10 / 70
▶ We need to get some numerical evidence on whether districts with low
STRs have higher test scores:
1. Compare average test scores in districts with low STRs to those with
high STRs (estimation)
2. Test the null hypothesis that the mean test scores in the two types of
districts are the same, against the alternative hypothesis that they
differ (hypothesis testing)
3. Estimate an interval for the difference in the mean test scores, high v.
low STR districts (confidence interval)
11 / 70
Initial data analysis
12 / 70
1. Estimation
̄ ̄ 1 𝑛 1 𝑛
𝑌𝑠𝑚𝑎𝑙𝑙 − 𝑌𝑙𝑎𝑟𝑔𝑒 = 𝑛𝑠𝑚𝑎𝑙𝑙 ∑𝑖=1
𝑠𝑚𝑎𝑙𝑙
𝑌𝑖 − 𝑛𝑙𝑎𝑟𝑔𝑒 ∑𝑖=1
𝑙𝑎𝑟𝑔𝑒
𝑌𝑖 (1)
= 657.4 − 650.0 (2)
= 7.4 (3)
13 / 70
2. Hypothesis testing
▶ Let’s test the null hypothesis that there is no difference between the
two groups
▶ The test statistic is
̄
𝑌𝑠𝑚𝑎𝑙𝑙 ̄
− 𝑌𝑙𝑎𝑟𝑔𝑒 ̄
𝑌𝑠𝑚𝑎𝑙𝑙 ̄
− 𝑌𝑙𝑎𝑟𝑔𝑒
𝑡= =
𝑠2𝑠𝑚𝑎𝑙𝑙 𝑠2𝑙𝑎𝑟𝑔𝑒 ̄ ̄
𝑆𝐸(𝑌𝑠𝑚𝑎𝑙𝑙 − 𝑌𝑙𝑎𝑟𝑔𝑒 )
√𝑛 + 𝑛𝑙𝑎𝑟𝑔𝑒
𝑠𝑚𝑎𝑙𝑙
̄
▶ where 𝑆𝐸(𝑌𝑠𝑚𝑎𝑙𝑙 ̄
− 𝑌𝑙𝑎𝑟𝑔𝑒 ) represents the standard error of
̄
𝑌𝑠𝑚𝑎𝑙𝑙 ̄
− 𝑌𝑙𝑎𝑟𝑔𝑒 and the variance estimates are calculated as
𝑛𝑠𝑚𝑎𝑙𝑙
1 2
𝑠2𝑠𝑚𝑎𝑙𝑙 = ∑ (𝑌𝑖 − 𝑌 ̄ )
𝑛𝑠𝑚𝑎𝑙𝑙 − 1 𝑖=1
14 / 70
2.
▶ |𝑡| > 1.96: reject the null hypothesis that the two means are the same
(at the 5% significance level)
15 / 70
3. Confidence interval
A 95% confidence interval for the difference between the means is,
16 / 70
R
library(dplyr)
caschool <- mutate(caschool, group = ifelse(str < 20, "small", "large"))
table <- group_by(caschool, group) %>%
summarize(Average = mean(testscr),
StdDev = sd(testscr),
N = n())
knitr::kable(table)
17 / 70
library(ggplot2)
ggplot(caschool, aes(group,testscr)) + geom_boxplot() + theme_bw()
testscr 690
660
630
large small
group
18 / 70
t.test(testscr ~ group, data = caschool)
19 / 70
What comes next
20 / 70
Review of Statistical Theory
21 / 70
Population, random variable, and distribution
▶ Population
▶ The group or collection of all possible entities of interest (school
districts)
▶ We will think of populations as infinitely large (∞ is an approximation
to “very big’ ’)
▶ Random variable Y
▶ Numerical summary of a random outcome (district average test score,
district STR)
22 / 70
Moments of a population distribution: mean, variance, standard
deviation, covariance, correlation
2. variance
▶ 𝐸(𝑌 –𝜇𝑌 )2 = 𝜎𝑌2
▶ measure of the squared spread of the distribution
23 / 70
𝐸(𝑌 −𝜇𝑌 )3
4. skewness = 𝜎𝑌3
𝐸(𝑌 −𝜇𝑌 )4
5. kurtosis: 𝜎𝑌4
24 / 70
25 / 70
2 random variables: joint distributions and covariance
26 / 70
▶ The covariance between Test Score and STR is negative:
c(cov(caschool$testscr, caschool$str),
cor(caschool$testscr, caschool$str))
27 / 70
The correlation coefficient is defined in terms of the covariance
28 / 70
▶ The correlation coefficient measures linear association
29 / 70
(c) Conditional distributions and conditional means
▶ Conditional distributions
▶ The distribution of Y, given value(s) of some other random variable, X
▶ Ex: the distribution of test scores, given that STR < 20
30 / 70
▶ Other examples of conditional means:
▶ Wages of all female workers (Y = wages, X = gender)
▶ Mortality rate of those given an experimental treatment (Y = live/die;
X = treated/not treated)
31 / 70
Distribution of a sample of data drawn randomly from a
population: 𝑌1 , ⋯ , 𝑌𝑛
32 / 70
Distribution of 𝑌1 , ⋯ , 𝑌𝑛 under simple random sampling
33 / 70
Estimation
34 / 70
Sampling distribution of 𝑌 ̄
35 / 70
▶ Example: Suppose Y takes on 0 or 1 (a Bernoulli random variable)
with probability distribution Pr(Y = 0) = 0.22, Pr(Y = 1) = 0.78
▶ Then
𝐸(𝑌 ) = 𝑝 ∗ 1 + (1 − 𝑝) ∗ 0 = 𝑝 = 0.78
𝜎 = 𝐸[(𝑌 − 𝜇𝑦 )2 ] = 𝑝 ∗ (1 − 𝑝)2 + (1 − 𝑝) ∗ (0 − 𝑝)2 = 𝑝(1 − 𝑝)
2
36 / 70
The sampling distribution of 𝑌 ̄ when Y is Bernoulli (p = 0.78):
37 / 70
R
6000
1500
4000
count
count 1000
2000
500
0 0
0.00 0.25 0.50 0.75 1.00 0.6 0.7 0.8 0.9
sample sample
38 / 70
▶ What is the mean of 𝑌 ̄ ? If 𝑌 ̄ = 𝜇 = 0.78, then 𝑌 ̄ is an unbiased
estimator of 𝜇
▶ What is the variance of 𝑌 ̄ ?
▶ How does 𝑌 ̄ depend on n
▶ Does 𝑌 ̄ become close to 𝜇 when n is large?
▶ Law of large numbers: 𝑌 ̄ is a consistent estimator of 𝜇
▶ 𝑌 ̄ − 𝜇 appears bell shaped for n large: is this generally true?
▶ In fact, 𝑌 ̄ − 𝜇 is approximately normally distributed for n large
(Central Limit Theorem)
39 / 70
The mean and variance of the sampling distribution of 𝑌 ̄
▶ In the general case that 𝑌𝑖 is i.i.d. from any distribution (not just
Bernoulli)
▶ Mean:
1 𝑁 1 𝑁 1 𝑁
𝐸(𝑌 ̄ ) = 𝐸 ( ∑ 𝑌𝑖 ) = ∑ 𝐸(𝑌𝑖 ) = ∑ 𝜇 = 𝜇𝑌
𝑁 𝑖=1 𝑁 𝑖=1 𝑁 𝑖=1 𝑌
▶ Variance:
2 2
𝑉 𝑎𝑟(𝑌 ̄ ) = 𝐸 [𝑌 ̄ − 𝐸(𝑌 ̄ )] = 𝐸 [𝑌 ̄ − 𝜇𝑌 ] =
2 2
1 𝑁 1 𝑁
=𝐸[ ∑ 𝑌𝑖 − 𝜇𝑌 ] = 𝐸 [ ∑(𝑌𝑖 − 𝜇𝑌 )] =
𝑁 𝑖=1 𝑁 𝑖=1
1 𝑁 𝑁
= ∑ ∑ 𝐸 [(𝑌𝑖 − 𝜇𝑌 )(𝑌𝑗 − 𝜇𝑌 )]
𝑁 2 𝑖=1 𝑗−1
1 𝑁 𝑁 1 𝑁 𝜎2
= 2
∑ ∑ 𝑐𝑜𝑣(𝑌𝑖 , 𝑌𝑗 ) = 2 ∑ 𝜎𝑌2 = 𝑌
𝑁 𝑖=1 𝑗−1 𝑁 𝑖=1 𝑁
40 / 70
The mean and variance of the sampling distribution of 𝑌 ̄
▶ MEAN: 𝐸(𝑌 ̄ ) = 𝜇𝑌
▶ 𝑌 ̄ is an unbiased estimator of 𝜇𝑌
2
𝜎𝑌
▶ VARIANCE: 𝑉 𝑎𝑟(𝑌 ̄ ) = 𝑁
▶ the spread of the sampling distribution is inversely related to 𝑁
▶ the larger the sample the more accurate the estimate
▶ As n increases, the distribution of 𝑌 ̄ becomes more tightly centered
around 𝜇𝑌 (Law of Large Numbers)
▶ Moreover, the distribution of 𝑌 ̄ − 𝜇𝑌 becomes normal (Central Limit
Theorem)
41 / 70
Law of Large Numbers (LLN)
𝑃 𝑟(|𝑌 ̄ − 𝜇𝑌 | < 𝜖) = 1 as n → ∞
𝑝
which can be written as 𝑌 ̄ −
→ 𝜇𝑌 (𝑌 ̄ converges in probability to 𝜇𝑌 )
42 / 70
Central Limit Theorem (CLT)
▶ [CLT] If 𝑌1 , ⋯ , 𝑌𝑁 are i.i.d. and 𝜎𝑌2 < ∞, then when n is large the
distribution of 𝑌 ̄ is well approximated by a normal distribution
𝜎𝑌2
𝑌 ̄ ∼ 𝑁 (𝜇𝑌 , )
𝑁
▶ and
√ 𝑌 ̄ − 𝜇𝑌
𝑛 ∼ 𝑁 (0, 1)
𝜎𝑌
▶ The larger n, the better the approximation
43 / 70
Sampling distribution of 𝑌 ̄ when Y is Bernoulli with p = 0.78:
44 / 70
▶ distribution of the standardized sample average (𝑌 ̄ − 𝐸(𝑌 ̄ ) /√𝑣𝑎𝑟(𝑌 ̄ )
45 / 70
R
for (i in 1:B){
Y = rbinom(N,1,p)
sample[i,1] = mean(Y, na.rm = T)
sample[i,2] = (sample[i,1] - p) / (p * (1-p) / N)^0.5 # here we are standardizing
}
library(ggplot2)
ggplot(sample, aes(std_average)) +
geom_histogram(aes(y = ..density..), binwidth = 0.2, fill="seagreen2") +
stat_function(fun = dnorm, args = list(mean = 0, sd = 1)) +
theme_classic()
}
46 / 70
set.seed(3453)
p = 0.78
N = 1000
B = 10000
myfunction(p,N,B)
0.4
0.3
density
0.2
0.1
0.0
−2.5 0.0 2.5
std_average
47 / 70
Summary: The Sampling Distribution of 𝑌 ̄
48 / 70
(b) Why Use 𝑌 ̄ To Estimate 𝜇𝑌 ?
▶ 𝑌 ̄ is unbiased: 𝐸(𝑌 ̄ ) = 𝜇𝑌
𝑝
▶ 𝑌 ̄ is consistent: 𝑌 ̄ −
→ 𝜇𝑌
▶ In addition: 𝑌 ̄ is the least squares estimator of 𝜇𝑌 , that is, it
minimizes the sum of squared errors (or residuals)
𝑛
2
𝑚𝑖𝑛𝑚 ∑ (𝑌𝑖 − 𝑚)
𝑖=1
49 / 70
Hypothesis Testing
▶ The hypothesis testing problem (for the mean): decide based on the
sample evidence whether
▶ a null hypothesis is true
▶ an alternative hypothesis is true
▶ In statistical notation:
▶ 𝐻0 ∶ 𝐸(𝑌 ) = 𝜇𝑌 ,0 vs. 𝐻1 ∶ 𝐸(𝑌 ) > 𝜇𝑌 ,0 (1-sided, >)
▶ 𝐻0 ∶ 𝐸(𝑌 ) = 𝜇𝑌 ,0 vs. 𝐻1 ∶ 𝐸(𝑌 ) < 𝜇𝑌 ,0 (1-sided, <)
▶ 𝐻0 ∶ 𝐸(𝑌 ) = 𝜇𝑌 ,0 vs. 𝐻1 ∶ 𝐸(𝑌 ) ≠ 𝜇𝑌 ,0 (2-sided)
50 / 70
▶ p-value = probability of drawing a statistic (e.g., 𝑌 ̄ ) at least as
adverse to the null as the value actually computed with your data,
assuming that the null hypothesis is true
▶ significance level: a pre-specified probability of incorrectly rejecting
the null, when the null is true
▶ Calculating the p-value based on 𝑌 ̄ :
51 / 70
▶ If n is large, the sampling distribution of 𝑌 ̄ is obtained using the
normal approximation (CLT):
𝑌 ̄ − 𝜇𝑌 ,0 𝑌 ̄ 𝑎𝑐𝑡𝑢𝑎𝑙 − 𝜇𝑌 ,0
= 𝑃 𝑟 [∣ ∣>∣ ∣]
𝜎𝑌 ̄ 𝜎𝑌 ̄
√
▶ where 𝜎𝑌 ̄ = 𝜎𝑌 / 𝑛
▶ Under normality, the p-value is calculated as the probability of values
𝑌 ̄ 𝑎𝑐𝑡𝑢𝑎𝑙 −𝜇𝑌 ,0
larger than ± 𝜎𝑌 ̄
on the left and right tails
52 / 70
Calculating the p-value with 𝜎𝑌 known
53 / 70
Estimator of the variance of Y
𝑛
1 2
𝑠2𝑌 = ∑ (𝑌 − 𝑌 ̄ )
𝑛 − 1 𝑖=1 𝑖
▶ if 𝑌𝑖 are i.i.d. then we can apply the LLN to 𝑠2𝑌 since it represents an
𝑝
average and be confident that 𝑠2𝑌 −
→ 𝜎𝑌2
𝑌 ̄ −𝜇𝑌 ,0
▶ Define 𝑡 = √ then the p-value is obtained as
𝑠𝑌 / 𝑁
54 / 70
What is the link between the p-value and the significance level?
55 / 70
▶ The value of 1.96 that we used to reject the null is the 5% critical
value
▶ The critical values of the Student t-distribution are tabulated in all
statistics books
▶ To reject the null hypothesis using the critical values:
1. Compute the t -statistic
2. Compute the degrees of freedom, which is 𝑛–1
3. Look up the 𝛼 critical value
4. The null is rejected if the t-statistic exceeds (in absolute value) this
critical value,
56 / 70
▶ The difference between the t-distribution and N(0,1) critical values
depend on the sample size n and becomes negligible as the sample is
larger than 120.
▶ Some 5% critical values for a 2-sided tests:
57 / 70
Figure 2: The four distributions of earnings are for women and men,
for those with only a high school diploma (a and c) and those whose
highest degree is from a four-year college (b and d).
58 / 70
▶ For n > 30 , the t-distribution and N(0,1) are very close and as n
grows large, the 𝑡𝑛–1 distribution converges to N(0,1)
▶ When sample sizes were small and there were no computers, the
t-distribution was providing correct inference
▶ For historical reasons, statistical software typically still use the
t-distribution to compute p-values, but this is irrelevant when the
sample size is moderate or large
▶ For these reasons, in this class we will focus on the large-n
approximation given by the CLT
59 / 70
Confidence Intervals
𝑠 𝑠
𝜇𝑌 ∈ (𝑌 ̄ − 1.96 √𝑌 , 𝑌 ̄ + 1.96 √𝑌 )
𝑛 𝑛
60 / 70
R
library(tidyquant)
sp <- tq_get("SPY", get = "stock.prices", from = "1995-01-01")
head(sp)
# A tibble: 6 x 8
symbol date open high low close volume adjusted
<chr> <date> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 SPY 1995-01-03 45.7 45.8 45.7 45.8 324300 28.3
2 SPY 1995-01-04 46.0 46 45.8 46 351800 28.4
3 SPY 1995-01-05 46.0 46.1 46.0 46 89800 28.4
4 SPY 1995-01-06 46.1 46.2 45.9 46.0 448400 28.5
5 SPY 1995-01-09 46.0 46.1 46 46.1 36800 28.5
6 SPY 1995-01-10 46.2 46.4 46.1 46.1 229800 28.5
tail(sp)
# A tibble: 6 x 8
symbol date open high low close volume adjusted
<chr> <date> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 SPY 2021-08-31 452. 452. 451. 452. 59300200 452.
2 SPY 2021-09-01 453. 453. 452. 452. 48721400 452.
3 SPY 2021-09-02 453. 454. 452. 453. 42501000 453.
4 SPY 2021-09-03 452. 454. 452. 453. 47170500 453.
5 SPY 2021-09-07 453. 453. 451. 451. 51671500 451.
6 SPY 2021-09-08 451. 452. 449. 451. 56121700 451.
61 / 70
ggplot(sp, aes(date, adjusted)) + ggplot(sp, aes(date, log(adjusted))) +
geom_line() + theme_bw() geom_line() + theme_bw()
6
400
log(adjusted)
300
adjusted
200
4
100
62 / 70
sp <- mutate(sp, return = 100*(log(adjusted) - log(lag(adjusted))))
tail(sp)
# A tibble: 6 x 9
symbol date open high low close volume adjusted return
<chr> <date> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 SPY 2021-08-31 452. 452. 451. 452. 59300200 452. -0.148
2 SPY 2021-09-01 453. 453. 452. 452. 48721400 452. 0.0531
3 SPY 2021-09-02 453. 454. 452. 453. 42501000 453. 0.307
4 SPY 2021-09-03 452. 454. 452. 453. 47170500 453. -0.0243
5 SPY 2021-09-07 453. 453. 451. 451. 51671500 451. -0.358
6 SPY 2021-09-08 451. 452. 449. 451. 56121700 451. -0.122
10
5
return
−5
−10
63 / 70
ggplot(sp, aes(return)) +
geom_histogram(aes(y = ..density..), binwidth = 0.1, fill = "steelblue1") +
stat_function(fun = dnorm,
args = list(mean = mean(sp$return,na.rm=T),
sd = sd(sp$return,na.rm=T))) +
theme_light()
0.6
0.4
density
0.2
0.0
−10 −5 0 5 10
return
64 / 70
library(psych)
c(skew(sp$return), kurtosi(sp$return))
describe(sp$return)
vars n mean sd median trimmed mad min max range skew kurtosis
X1 1 6718 0.04 1.21 0.07 0.07 0.79 -11.59 13.56 25.15 -0.3 11.29
se
X1 0.01
65 / 70
fin.data <- tq_get(c("AAPL", "BAC", "SPY"), get = "stock.prices", from = "1995-
head(fin.data)
# A tibble: 6 x 8
symbol date open high low close volume adjusted
<chr> <date> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 AAPL 1995-01-03 0.347 0.347 0.338 0.343 103868800 0.291
2 AAPL 1995-01-04 0.345 0.354 0.345 0.352 158681600 0.298
3 AAPL 1995-01-05 0.350 0.352 0.346 0.347 73640000 0.295
4 AAPL 1995-01-06 0.372 0.385 0.367 0.375 1076622400 0.318
5 AAPL 1995-01-09 0.372 0.374 0.366 0.368 274086400 0.312
6 AAPL 1995-01-10 0.368 0.393 0.368 0.390 614790400 0.331
tail(fin.data)
# A tibble: 6 x 8
symbol date open high low close volume adjusted
<chr> <date> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 SPY 2021-08-31 452. 452. 451. 452. 59300200 452.
2 SPY 2021-09-01 453. 453. 452. 452. 48721400 452.
3 SPY 2021-09-02 453. 454. 452. 453. 42501000 453.
4 SPY 2021-09-03 452. 454. 452. 453. 47170500 453.
5 SPY 2021-09-07 453. 453. 451. 451. 51671500 451.
6 SPY 2021-09-08 451. 452. 449. 451. 56121700 451.
66 / 70
ret.data = fin.data %>%
group_by(symbol) %>%
mutate(return = 100 * (log(adjusted) - log(lag(adjusted)))) %>%
select(symbol, date, return)
head(ret.data)
# A tibble: 6 x 3
# Groups: symbol [1]
symbol date return
<chr> <date> <dbl>
1 AAPL 1995-01-03 NA
2 AAPL 1995-01-04 2.57
3 AAPL 1995-01-05 -1.28
4 AAPL 1995-01-06 7.73
5 AAPL 1995-01-09 -1.92
6 AAPL 1995-01-10 5.85
ret.data %>%
group_by(symbol) %>%
summarize(Average = mean(return, na.rm = T),
StdDev = sd(return, na.rm = T),
N = n()) %>%
knitr::kable()
67 / 70
library(tidyr)
ret.data <- ret.data %>%
pivot_wider(names_from = symbol, values_from = return)
head(ret.data)
# A tibble: 6 x 4
date AAPL BAC SPY
<date> <dbl> <dbl> <dbl>
1 1995-01-03 NA NA NA
2 1995-01-04 2.57 0.816 0.477
3 1995-01-05 -1.28 0.810 0
4 1995-01-06 7.73 -0.269 0.102
5 1995-01-09 -1.92 0 0.102
6 1995-01-10 5.85 1.60 0.102
68 / 70
▶ 𝐻0 ∶ 𝜇𝑆𝑃 𝑌 = 0 vs 𝐻1 ∶ 𝜇𝑆 𝑃 𝑌 ≠ 0
▶ t = …..
▶ Reject or fail to reject?
69 / 70
Summary
70 / 70