1. Why we conduct a t test?
- A t-test is a statistical test that is used to compare the means of two groups.
- It is often used in hypothesis testing to determine whether a process or treatment actually has an effect on the population of interest, or whether two groups are different each other.
- The Central Limit Theorem suggests that even if the original variables themselves are not normally distributed, distribution of samples converges to normal as the number of samples increases.
→ This is the theoretical foundation that we can conduct statistical inference and hypothesis test using a normal distribution.
- However, it is not always the case that we can have access to sizable data.
- t test provides us a solution of this small N problem in hypothesis testing.
2. What type of t-test should we use?
When you conduct a t-test, you need to consider two things:
(1) whether the groups being compared come from a single population or two different populations
(3) whether you want to test the difference in a specific direction or both directions
One-sample, two-sample, or paired t-test?
- If the groups come from a single population (for example, measuring before and after an medical treatment), perform a paired t-test.
- If the groups come from two different populations (for exmaple, comparing which hamberger tastes better between McDonald’s and In-N-Out Burger), perform a two-sample t-test.
- If there is one group being compared against a standard value (for example, comparing your test score to the averaged test score of your friends in the same class), perform a one-sample t-test.
One-tailed or two-tailed t-test?
- If you only care whether the two populations are different from one another, perform a two-tailed t-test.
- If you want to know whether one population mean is greater than or less than the other, perform a one-tailed t-test.
- A one-tailed t-test is way harder to be passed.
3. Statistical significance
significance level |
α |
The probability of the study rejecting the null hypothesis |
significance probability |
p-value |
The probability of obtaining a result at least as extreme, given that the null hypothesis is true. |
|
|
|
- In statistical hypothesis testing, a result has statistical significance when it is very unlikely to have occurred given the null hypothesis.
- The null hypothesis (often denoted \(H_0\) ) is a default hypothesis that a quantity to be measured is zero (null).
- The alternative hypothesis (often denoted \(H_1\) ) is a position that states something is happening, a new theory is preferred instead of an old one (null hypothesis).
- A study’s defined significance level, denoted by \(α\), is the probability of the study rejecting the null hypothesis, given that the null hypothesis was assumed to be true.
- Significance probability, the p-value of a result (\(p\)), is the probability of obtaining a result at least as extreme, given that the null hypothesis is true.
- The result is statistically significant, by the standards of the study, when \(p≤α\).
- The significance level for a study is chosen before data collection, and is typically set to 5% or much lower—depending on the field of study.
3.1 Process of t test
- Let’s suppose you drew 10 numbers (1, 2, 3, 4, 5, 6, 7, 8, 9, 10) from the population (population mean \(μ\) = 5.5)
- The sample mean = 5.5
(1, 2, 3, 4, 5, 6, 7, 8, 9, 10)/10 = 5.5
- What we want to know:
Whether or not the population mean is 5.
- To confirm this, we need to conduct t test.
Process of t test
Show null hypothesis
: \(H_0\)
Show alternative hypothesis
: \(H_1\)
Calculate t value
Identify the critical values
Check if your t value
is within the rejection area
Conclusion
Process of t test
in this case
- \(H_0\): the population mean = 5
- What we want to know is “The sample mean is 5.5. With this result, can we conclude that the population mean is 5?”
- \(H_1\): the population mean is not 5
- \(H_0\) and \(H_1\) are
mutually exclusive
- Calculate \(t value\) with the following equation:
\[T = \frac{\bar{x} - μ_0}{SE} = \frac{\bar{x} - μ_0}{u_x / \sqrt{n}}\]
- \(\bar{x}\) : Sample mean (= 5.5)
- \(μ_0\) : The value we want to estimate (= 5)
- \(n\) : Sample size (= 10)
- \(u_x\): unbiased standard deviation
- \(SE\) :
standard Error: SE
★ \(u_x^2\) can be calculated with the following eauation:
\[u_x^2 = \frac{\sum_{i=1}^n (x_i - \bar{x})^2}{n-1}\]
- \(u_x\) = 3.03
- If we plug in \(μ_0\) = 5 in the equation above, we get the following t value:
\[T = \frac{\bar{x} - μ_0}{u_x / \sqrt{n}}\]
\[ = \frac{{5.5} - 5}{3.03 / \sqrt{10}}\]
\[ = 0.522\]
- We want to know whether or not the population mean is 5.
Point estimation
- Point estimation involves the use of sample data to calculate a single value (known as a point estimate) which is to serve as a “best guess” or “best estimate” of an unknown population parameter (for example, the population mean).
- We estimate the sample mean, 0.522, and infer whether the population mean is 5.
- Numbers in each cell is the critical value for 7 levels of
significance level
.
- We reject the null hypothesis when the t value we get from our sample (0.522 in this case) is larger than the critical value (which we are trying to identify here).
- \(v\) is the
degree of freedom
: the number of sample size - 1
→ 9 in this case (= 10 - 1)
- \(α\) is
significance level
:
- Usually we use the significance level \((α = 0.05)\) for one-tailed t-test.
→ We use the significance level \((α = 0.025)\) for two-tailed t-test here.
- See the intersection of \((α = 0.025)\) & \(v=9\) in the table
- The value 2.262 is the critical value
- If the t value we get from our sample (0.05) lies between -2.26 and 2.26, then we fail to reject the null hypothesis.
- If the t value we get from our sample (0.05) lies out of this range, then we reject the null hypothesis.
- 0.05 lies between -2.26 and 2.26
→ we cannot reject the null hypothesis : population mean is 5
→ Based on the sample we get, the population mean CAN BE 5.
3.3 t test
on R
- We draw the 10 values from a population, 1, 2, …, 10
- We do not know its population mean
score <- c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10)
- We want to know if
the population mean = 5
or not
- Null hypothesis:
the population mean = 5
One Sample t-test
data: score
t = 0.52223, df = 9, p-value = 0.6141
alternative hypothesis: true mean is not equal to 5
95 percent confidence interval:
3.334149 7.665851
sample estimates:
mean of x
5.5
- See the row 3:
- \(t\) = 0.52223,
p-value
= 0.6141
p-value
= 0.6141 means the p-value
in a two-tailed ttest
- See the row 4:
alternative hypothesis: true mean is not equal to 5
p-value
(0.6141) is larger than 0.05
→ We cannot reject the null hypothesis: the population mean = 5
→ Based on the sample we get, the population mean CAN BE 5.
3.5 EXERCISE
- We conducted a survey to 10 Waseda students who bought cell phones.
- The followings are the information we got.
- The averaged price of a cell phone is 65,000 yen
- Standard deviation = 10,000 yen.
Question
: Can you say that the population mean of the price of a cell phone is 50,000 yen? Calculate t value both by hand and on R can check if you get the same results.
3.6 EXERCISE
- Tom opened a case near the Waseda campus, named Printemps.
- Total sales for the last 8 days are the followings:
45, 39, 42, 57, 28, 33, 40, 52(unit: 10,000 yen)
Question
: Can you say that the population mean of the cafe Printems is 500,000 yen? Calculate t value both by hand and on R can check if you get the same results.
4. Estimation on Population Propotion
内閣支持29%、不支持52%(NHK世論調査)記事
- This article reports the result of a survey conducted by NHK (August 10, 2021)
- The survey was conducted on August 2021.
- Cabinet support rate for the Suga Administration is 29%
- N = 1214
Question
: Can you say that the cabinet rate for the Suga Administration is less than 30%?
- This is similar to ttest, but we should not use it.
- We need to use proportion test on R:
prop.test() function
- Since 29% of 12114 respondents support the Suga Administration, 352 of 1214 supported the Suga Administration
prop.test(c(352), c(1214))
1-sample proportions test with continuity correction
data: c(352) out of c(1214), null probability 0.5
X-squared = 213.41, df = 1, p-value < 2.2e-16
alternative hypothesis: true p is not equal to 0.5
95 percent confidence interval:
0.2647212 0.3165265
sample estimates:
p
0.2899506
Null hypothesis
: the cabinet support rate in population = 0.5 (50%)
→ p-value < 2.2e-16 (= 0.00000000000000022)
- We can reject the null hypothesis
→ The cabinet support rate in population is not 50%
- See
95 percent confidence interval: 0.2647212 --- 0.3165265
→ The true cabinet support rate in population must be somewhere between 26.47% ~ 31.66%
- 30% is within the range of 95% confidence interval.
- We cannot say that the cabinet rate for the Suga Administration is less than 30%.
What if 300 respondents supported the Suga administration?
prop.test(c(300), c(1214))
1-sample proportions test with continuity correction
data: c(300) out of c(1214), null probability 0.5
X-squared = 309.53, df = 1, p-value < 2.2e-16
alternative hypothesis: true p is not equal to 0.5
95 percent confidence interval:
0.2232793 0.2725771
sample estimates:
p
0.247117
- See
95 percent confidence interval: 0.2232793 --- 0.2725771
→ The true cabinet support rate in population must be smaller than 30%.
- 30% is out of the range of 95% confidence interval.
- We can say that the cabinet rate for the Suga Administration is less than 30%.
参考文献
宋財泫 (Jaehyun Song)・矢内勇生 (Yuki Yanai)「私たちのR: ベストプラクティスの探究」
土井翔平(北海道大学公共政策大学院)「Rで計量政治学入門」
矢内勇生(高知工科大学)授業一覧
浅野正彦, 矢内勇生.『Rによる計量政治学』オーム社、2018年
浅野正彦, 中村公亮.『初めてのRStudio』オーム社、2018年
Winston Chang, R Graphics Cookbook, O’Reilly Media, 2012.
Kieran Healy, DATA VISUALIZATION, Princeton, 2019
Kosuke Imai, Quantitative Social Science: An Introduction, Princeton University Press, 2017