|
|
|
1 |
|
00:00:11,850 --> 00:00:16,370 |
|
Inshallah we'll start numerical descriptive measures |
|
|
|
2 |
|
00:00:16,370 --> 00:00:22,270 |
|
for the population. Last time we talked about the |
|
|
|
3 |
|
00:00:22,270 --> 00:00:25,780 |
|
same measures. I mean the same descriptive measures |
|
|
|
4 |
|
00:00:25,780 --> 00:00:29,180 |
|
for a sample. And we have already talked about the |
|
|
|
5 |
|
00:00:29,180 --> 00:00:35,080 |
|
mean, variance, and standard deviation. These are |
|
|
|
6 |
|
00:00:35,080 --> 00:00:38,580 |
|
called statistics because they are computed from |
|
|
|
7 |
|
00:00:38,580 --> 00:00:43,140 |
|
the sample. Here we'll see how can we do the same |
|
|
|
8 |
|
00:00:44,300 --> 00:00:47,320 |
|
measures but for a population, I mean for the |
|
|
|
9 |
|
00:00:47,320 --> 00:00:53,020 |
|
entire dataset. So descriptive statistics |
|
|
|
10 |
|
00:00:53,020 --> 00:00:57,860 |
|
described previously in the last two lectures was |
|
|
|
11 |
|
00:00:57,860 --> 00:01:04,200 |
|
for a sample. Here we'll just see how can we |
|
|
|
12 |
|
00:01:04,200 --> 00:01:07,740 |
|
compute these measures for the entire population. |
|
|
|
13 |
|
00:01:08,480 --> 00:01:11,600 |
|
In this case, the statistics we talked about |
|
|
|
14 |
|
00:01:11,600 --> 00:01:17,200 |
|
before are called And if you remember the first |
|
|
|
15 |
|
00:01:17,200 --> 00:01:19,800 |
|
lecture, we said there is a difference between |
|
|
|
16 |
|
00:01:19,800 --> 00:01:24,300 |
|
statistics and parameters. A statistic is a value |
|
|
|
17 |
|
00:01:24,300 --> 00:01:27,520 |
|
that computed from a sample, but parameter is a |
|
|
|
18 |
|
00:01:27,520 --> 00:01:32,140 |
|
value computed from population. So the important |
|
|
|
19 |
|
00:01:32,140 --> 00:01:37,020 |
|
population parameters are population mean, |
|
|
|
20 |
|
00:01:37,660 --> 00:01:43,560 |
|
variance, and standard deviation. Let's start with |
|
|
|
21 |
|
00:01:43,560 --> 00:01:45,880 |
|
the first one, the mean, or the population mean. |
|
|
|
22 |
|
00:01:46,980 --> 00:01:50,720 |
|
As the sample mean is defined by the sum of the |
|
|
|
23 |
|
00:01:50,720 --> 00:01:55,120 |
|
values divided by the sample size. But here, we |
|
|
|
24 |
|
00:01:55,120 --> 00:01:57,880 |
|
have to divide by the population size. So that's |
|
|
|
25 |
|
00:01:57,880 --> 00:02:01,140 |
|
the difference between sample mean and population |
|
|
|
26 |
|
00:02:01,140 --> 00:02:08,950 |
|
mean. For the sample mean, we use x bar. Here we |
|
|
|
27 |
|
00:02:08,950 --> 00:02:14,790 |
|
use Greek letter, mu. This is pronounced as mu. So |
|
|
|
28 |
|
00:02:14,790 --> 00:02:18,790 |
|
mu is the sum of the x values divided by the |
|
|
|
29 |
|
00:02:18,790 --> 00:02:21,210 |
|
population size, not the sample size. So it's |
|
|
|
30 |
|
00:02:21,210 --> 00:02:24,570 |
|
quite similar to the sample mean. So mu is the |
|
|
|
31 |
|
00:02:24,570 --> 00:02:28,030 |
|
population mean, n is the population size, and xi |
|
|
|
32 |
|
00:02:28,030 --> 00:02:33,270 |
|
is the ith value of the variable x. Similarly, for |
|
|
|
33 |
|
00:02:33,270 --> 00:02:37,310 |
|
the other parameter, which is the variance, the |
|
|
|
34 |
|
00:02:37,310 --> 00:02:41,520 |
|
variance There is a little difference between the |
|
|
|
35 |
|
00:02:41,520 --> 00:02:45,480 |
|
sample and population variance. Here, we subtract |
|
|
|
36 |
|
00:02:45,480 --> 00:02:49,700 |
|
the population mean instead of the sample mean. So |
|
|
|
37 |
|
00:02:49,700 --> 00:02:55,140 |
|
sum of xi minus mu squared, then divide by this |
|
|
|
38 |
|
00:02:55,140 --> 00:02:59,140 |
|
population size, capital N, instead of N minus 1. |
|
|
|
39 |
|
00:02:59,520 --> 00:03:02,260 |
|
So that's the difference between sample and |
|
|
|
40 |
|
00:03:02,260 --> 00:03:07,020 |
|
population variance. So again, in the sample |
|
|
|
41 |
|
00:03:07,020 --> 00:03:12,080 |
|
variance, we subtracted x bar. Here, we subtract |
|
|
|
42 |
|
00:03:12,080 --> 00:03:15,640 |
|
the mean of the population, mu, then divide by |
|
|
|
43 |
|
00:03:15,640 --> 00:03:20,200 |
|
capital N instead of N minus 1. So the |
|
|
|
44 |
|
00:03:20,200 --> 00:03:24,000 |
|
computations for the sample and the population |
|
|
|
45 |
|
00:03:24,000 --> 00:03:30,220 |
|
mean or variance are quite similar. Finally, the |
|
|
|
46 |
|
00:03:30,220 --> 00:03:35,390 |
|
population standard deviation. is the same as the |
|
|
|
47 |
|
00:03:35,390 --> 00:03:38,810 |
|
sample population variance and here just take the |
|
|
|
48 |
|
00:03:38,810 --> 00:03:43,170 |
|
square root of the population variance and again |
|
|
|
49 |
|
00:03:43,170 --> 00:03:47,170 |
|
as we did as we explained before the standard |
|
|
|
50 |
|
00:03:47,170 --> 00:03:51,550 |
|
deviation has the same units as the original unit |
|
|
|
51 |
|
00:03:51,550 --> 00:03:57,130 |
|
so nothing is new we just extend the sample |
|
|
|
52 |
|
00:03:57,130 --> 00:04:02,410 |
|
statistic to the population parameter and again |
|
|
|
53 |
|
00:04:04,030 --> 00:04:08,790 |
|
The mean is denoted by mu, it's a Greek letter. |
|
|
|
54 |
|
00:04:10,210 --> 00:04:12,790 |
|
The population variance is denoted by sigma |
|
|
|
55 |
|
00:04:12,790 --> 00:04:17,030 |
|
squared. And finally, the population standard |
|
|
|
56 |
|
00:04:17,030 --> 00:04:21,130 |
|
deviation is denoted by sigma. So that's the |
|
|
|
57 |
|
00:04:21,130 --> 00:04:24,250 |
|
numerical descriptive measures either for a sample |
|
|
|
58 |
|
00:04:24,250 --> 00:04:28,590 |
|
or a population. So just summary for these |
|
|
|
59 |
|
00:04:28,590 --> 00:04:33,330 |
|
measures. The measures are mean variance, standard |
|
|
|
60 |
|
00:04:33,330 --> 00:04:38,250 |
|
deviation. Population parameters are mu for the |
|
|
|
61 |
|
00:04:38,250 --> 00:04:43,830 |
|
mean, sigma squared for variance, and sigma for |
|
|
|
62 |
|
00:04:43,830 --> 00:04:46,710 |
|
standard deviation. On the other hand, for the |
|
|
|
63 |
|
00:04:46,710 --> 00:04:51,430 |
|
sample statistics, we have x bar for sample mean, |
|
|
|
64 |
|
00:04:52,110 --> 00:04:56,750 |
|
s squared for the sample variance, and s is the |
|
|
|
65 |
|
00:04:56,750 --> 00:05:00,410 |
|
sample standard deviation. That's sample |
|
|
|
66 |
|
00:05:00,410 --> 00:05:05,360 |
|
statistics against population parameters. Any |
|
|
|
67 |
|
00:05:05,360 --> 00:05:05,700 |
|
question? |
|
|
|
68 |
|
00:05:10,940 --> 00:05:17,240 |
|
Let's move to a new topic, which is empirical rule. |
|
|
|
69 |
|
00:05:19,340 --> 00:05:25,620 |
|
Now, empirical rule is just we |
|
|
|
70 |
|
00:05:25,620 --> 00:05:30,120 |
|
have to approximate the variation of data in case |
|
|
|
71 |
|
00:05:30,120 --> 00:05:34,950 |
|
of They'll shift. I mean suppose the data is |
|
|
|
72 |
|
00:05:34,950 --> 00:05:37,770 |
|
symmetric around the mean. I mean by symmetric |
|
|
|
73 |
|
00:05:37,770 --> 00:05:42,310 |
|
around the mean, the mean is the vertical line |
|
|
|
74 |
|
00:05:42,310 --> 00:05:46,570 |
|
that splits the data into two halves. One to the |
|
|
|
75 |
|
00:05:46,570 --> 00:05:49,570 |
|
right and the other to the left. I mean, the mean, |
|
|
|
76 |
|
00:05:49,870 --> 00:05:52,650 |
|
the area to the right of the mean equals 50%, |
|
|
|
77 |
|
00:05:52,650 --> 00:05:54,970 |
|
which is the same as the area to the left of the |
|
|
|
78 |
|
00:05:54,970 --> 00:05:58,710 |
|
mean. Now suppose or consider the data is bell |
|
|
|
79 |
|
00:05:58,710 --> 00:06:02,570 |
|
-shaped. Bell-shaped, normal, or symmetric? So |
|
|
|
80 |
|
00:06:02,570 --> 00:06:04,290 |
|
it's not skewed either to the right or to the |
|
|
|
81 |
|
00:06:04,290 --> 00:06:08,030 |
|
left. So here we assume, okay, the data is bell |
|
|
|
82 |
|
00:06:08,030 --> 00:06:13,430 |
|
-shaped. In this scenario, in this case, there is |
|
|
|
83 |
|
00:06:13,430 --> 00:06:22,100 |
|
a rule called 68, 95, 99.7 rule. Number one, |
|
|
|
84 |
|
00:06:22,960 --> 00:06:26,300 |
|
approximately 68% of the data in a bell-shaped |
|
|
|
85 |
|
00:06:26,300 --> 00:06:31,780 |
|
lies within one standard deviation of the |
|
|
|
86 |
|
00:06:31,780 --> 00:06:37,100 |
|
population. So this is the first rule, 68% of the |
|
|
|
87 |
|
00:06:37,100 --> 00:06:43,920 |
|
data or of the observations Lie within a mu minus |
|
|
|
88 |
|
00:06:43,920 --> 00:06:48,880 |
|
sigma and a mu plus sigma. That's the meaning of |
|
|
|
89 |
|
00:06:48,880 --> 00:06:51,800 |
|
the data in a bell-shaped distribution is within one |
|
|
|
90 |
|
00:06:51,800 --> 00:06:55,900 |
|
standard deviation of mean or mu plus or minus |
|
|
|
91 |
|
00:06:55,900 --> 00:07:01,480 |
|
sigma. So again, you can say that if the data is |
|
|
|
92 |
|
00:07:01,480 --> 00:07:04,100 |
|
normally distributed or if the data is bell |
|
|
|
93 |
|
00:07:04,100 --> 00:07:12,210 |
|
shaped, that is 68% of the data lies within one |
|
|
|
94 |
|
00:07:12,210 --> 00:07:16,250 |
|
standard deviation of the mean, either below or |
|
|
|
95 |
|
00:07:16,250 --> 00:07:21,710 |
|
above it. So 68% of the data. So this is the first |
|
|
|
96 |
|
00:07:21,710 --> 00:07:22,090 |
|
rule. |
|
|
|
97 |
|
00:07:29,050 --> 00:07:37,170 |
|
68% of the data lies between mu minus sigma and mu |
|
|
|
98 |
|
00:07:37,170 --> 00:07:37,750 |
|
plus sigma. |
|
|
|
99 |
|
00:07:40,480 --> 00:07:46,260 |
|
The other rule is approximately 95% of the data in |
|
|
|
100 |
|
00:07:46,260 --> 00:07:48,980 |
|
a bell-shaped distribution lies within two |
|
|
|
101 |
|
00:07:48,980 --> 00:07:53,240 |
|
standard deviations of the mean. That means this |
|
|
|
102 |
|
00:07:53,240 --> 00:08:00,880 |
|
area covers between minus two sigma and plus mu |
|
|
|
103 |
|
00:08:00,880 --> 00:08:08,360 |
|
plus two sigma. So 95% of the data lies between |
|
|
|
104 |
|
00:08:08,360 --> 00:08:15,410 |
|
minus mu two sigma And finally, |
|
|
|
105 |
|
00:08:15,790 --> 00:08:21,270 |
|
approximately 99.7% of the data, it means almost |
|
|
|
106 |
|
00:08:21,270 --> 00:08:25,490 |
|
the data. Because we are saying 99.7 means most of |
|
|
|
107 |
|
00:08:25,490 --> 00:08:29,930 |
|
the data falls or lies within three standard |
|
|
|
108 |
|
00:08:29,930 --> 00:08:37,770 |
|
deviations of the mean. So 99.7% of the data lies |
|
|
|
109 |
|
00:08:37,770 --> 00:08:41,470 |
|
between mu minus the pre-sigma and the mu plus of |
|
|
|
110 |
|
00:08:41,470 --> 00:08:41,870 |
|
pre-sigma. |
|
|
|
111 |
|
00:08:45,030 --> 00:08:49,810 |
|
68, 95, 99.7 are fixed numbers. Later in chapter |
|
|
|
112 |
|
00:08:49,810 --> 00:08:55,010 |
|
6, we will explain in details other coefficients. |
|
|
|
113 |
|
00:08:55,530 --> 00:08:58,250 |
|
Maybe suppose we are interested not in one of |
|
|
|
114 |
|
00:08:58,250 --> 00:09:03,010 |
|
these. Suppose we are interested in 90% or 80% or |
|
|
|
115 |
|
00:09:03,010 --> 00:09:11,500 |
|
85%. This rule just for 689599.7. This rule is |
|
|
|
116 |
|
00:09:11,500 --> 00:09:15,560 |
|
called 689599 |
|
|
|
117 |
|
00:09:15,560 --> 00:09:22,960 |
|
.7 rule. That is, again, 68% of the data lies |
|
|
|
118 |
|
00:09:22,960 --> 00:09:27,030 |
|
within one standard deviation of the mean. 95% of |
|
|
|
119 |
|
00:09:27,030 --> 00:09:30,370 |
|
the data lies within two standard deviations of |
|
|
|
120 |
|
00:09:30,370 --> 00:09:33,850 |
|
the mean. And finally, most of the data falls |
|
|
|
121 |
|
00:09:33,850 --> 00:09:36,950 |
|
within three standard deviations of the mean. |
|
|
|
122 |
|
00:09:39,870 --> 00:09:43,330 |
|
Let's see how can we use this empirical rule for a |
|
|
|
123 |
|
00:09:43,330 --> 00:09:49,850 |
|
specific example. Imagine that the variable math |
|
|
|
124 |
|
00:09:49,850 --> 00:09:54,070 |
|
test scores is bell shaped. So here we assume that |
|
|
|
125 |
|
00:09:55,230 --> 00:10:00,950 |
|
The math test score has symmetric shape or bell |
|
|
|
126 |
|
00:10:00,950 --> 00:10:04,230 |
|
shape. In this case, we can use the previous rule. |
|
|
|
127 |
|
00:10:04,350 --> 00:10:09,610 |
|
Otherwise, we cannot. So assume the math test |
|
|
|
128 |
|
00:10:09,610 --> 00:10:15,750 |
|
score is bell-shaped with a mean of 500. I mean, |
|
|
|
129 |
|
00:10:16,410 --> 00:10:19,750 |
|
the population mean is 500 and standard deviation |
|
|
|
130 |
|
00:10:19,750 --> 00:10:24,620 |
|
of 90. And let's see how can we apply the |
|
|
|
131 |
|
00:10:24,620 --> 00:10:29,220 |
|
empirical rule. So again, meta score has a mean of |
|
|
|
132 |
|
00:10:29,220 --> 00:10:35,300 |
|
500 and standard deviation sigma is 90. Then we |
|
|
|
133 |
|
00:10:35,300 --> 00:10:43,200 |
|
can say that 60% of all test takers scored between |
|
|
|
134 |
|
00:10:43,200 --> 00:10:46,640 |
|
68%. |
|
|
|
135 |
|
00:10:46,640 --> 00:10:56,550 |
|
So mu is 500. minus sigma is 90. And mu plus |
|
|
|
136 |
|
00:10:56,550 --> 00:11:05,390 |
|
sigma, 500 plus 90. So you can say that 68% or 230 |
|
|
|
137 |
|
00:11:05,390 --> 00:11:15,610 |
|
of all test takers scored between 410 and 590. So |
|
|
|
138 |
|
00:11:15,610 --> 00:11:22,900 |
|
68% of all test takers who took that exam scored |
|
|
|
139 |
|
00:11:22,900 --> 00:11:27,740 |
|
between 14 and 590. That if we assume previously |
|
|
|
140 |
|
00:11:27,740 --> 00:11:29,980 |
|
the data is well shaped, otherwise we cannot say |
|
|
|
141 |
|
00:11:29,980 --> 00:11:36,420 |
|
that. For the other rule, 95% of all test takers |
|
|
|
142 |
|
00:11:36,420 --> 00:11:44,400 |
|
scored between mu is 500 minus 2 times sigma, 500 |
|
|
|
143 |
|
00:11:44,400 --> 00:11:49,760 |
|
plus 2 times sigma. So that means 500 minus 180 is |
|
|
|
144 |
|
00:11:49,760 --> 00:11:55,100 |
|
320. 500 plus 180 is 680. So you can say that |
|
|
|
145 |
|
00:11:55,100 --> 00:11:59,080 |
|
approximately 95% of all test takers scored |
|
|
|
146 |
|
00:11:59,080 --> 00:12:07,860 |
|
between 320 and 680. Finally, you can say that |
|
|
|
147 |
|
00:12:10,770 --> 00:12:13,570 |
|
all of the test takers, approximately all, because |
|
|
|
148 |
|
00:12:13,570 --> 00:12:20,030 |
|
when we are saying 99.7 it means just 0.3 is the |
|
|
|
149 |
|
00:12:20,030 --> 00:12:23,590 |
|
rest, so you can say approximately all test takers |
|
|
|
150 |
|
00:12:23,590 --> 00:12:30,730 |
|
scored between mu minus three sigma which is 90 |
|
|
|
151 |
|
00:12:30,730 --> 00:12:39,830 |
|
and mu It lost 3 seconds. So 500 minus 3 times 9 |
|
|
|
152 |
|
00:12:39,830 --> 00:12:45,950 |
|
is 270. So that's 230. 500 plus 270 is 770. So we |
|
|
|
153 |
|
00:12:45,950 --> 00:12:49,690 |
|
can say that 99.7% of all the stackers scored |
|
|
|
154 |
|
00:12:49,690 --> 00:12:55,610 |
|
between 230 and 770. I will give another example |
|
|
|
155 |
|
00:12:55,610 --> 00:12:59,210 |
|
just to make sure that you understand the meaning |
|
|
|
156 |
|
00:12:59,210 --> 00:13:00,870 |
|
of this rule. |
|
|
|
157 |
|
00:13:03,620 --> 00:13:09,720 |
|
For business, a statistic goes. |
|
|
|
158 |
|
00:13:15,720 --> 00:13:20,720 |
|
For business, a statistic example. Suppose the |
|
|
|
159 |
|
00:13:20,720 --> 00:13:29,740 |
|
scores are bell-shaped. So we are assuming the |
|
|
|
160 |
|
00:13:29,740 --> 00:13:40,970 |
|
data is bell-shaped. with a mean of 75 and standard |
|
|
|
161 |
|
00:13:40,970 --> 00:13:41,950 |
|
deviation of 5. |
|
|
|
162 |
|
00:13:44,990 --> 00:13:53,810 |
|
Also, let's assume that 100 students took |
|
|
|
163 |
|
00:13:53,810 --> 00:14:00,840 |
|
the exam. So we have 100 students. Last year took |
|
|
|
164 |
|
00:14:00,840 --> 00:14:05,360 |
|
the exam of business statistics. The mean was 75. |
|
|
|
165 |
|
00:14:06,240 --> 00:14:10,920 |
|
And standard deviation was 5. And let's see how it |
|
|
|
166 |
|
00:14:10,920 --> 00:14:17,100 |
|
can tell about the 68% rule. It means that 68% |
|
|
|
167 |
|
00:14:17,100 --> 00:14:22,100 |
|
of all the students scored |
|
|
|
168 |
|
00:14:22,100 --> 00:14:28,650 |
|
between mu minus sigma. Mu is 75. minus sigma and |
|
|
|
169 |
|
00:14:28,650 --> 00:14:29,610 |
|
the mu plus sigma. |
|
|
|
170 |
|
00:14:33,590 --> 00:14:39,290 |
|
So that means 68 students, because we have 100, so |
|
|
|
171 |
|
00:14:39,290 --> 00:14:45,410 |
|
you can say 68 students scored between 70 and 80. |
|
|
|
172 |
|
00:14:46,610 --> 00:14:53,290 |
|
So 68 students out of 100 scored between 70 and |
|
|
|
173 |
|
00:14:53,290 --> 00:15:02,990 |
|
80. About 95 students out of 100 scored between 75 |
|
|
|
174 |
|
00:15:02,990 --> 00:15:12,190 |
|
minus 2 times 5. 75 plus 2 times 5. So that gives |
|
|
|
175 |
|
00:15:12,190 --> 00:15:13,770 |
|
65. |
|
|
|
176 |
|
00:15:15,550 --> 00:15:20,950 |
|
The minimum and the maximum is 85. So you can say |
|
|
|
177 |
|
00:15:20,950 --> 00:15:25,930 |
|
that around 95 students scored between 65 and 85. |
|
|
|
178 |
|
00:15:26,650 --> 00:15:33,510 |
|
Finally, maybe you can see all students. Because |
|
|
|
179 |
|
00:15:33,510 --> 00:15:38,650 |
|
when you're saying 99.7, it means almost all the |
|
|
|
180 |
|
00:15:38,650 --> 00:15:47,210 |
|
students scored between 75 minus 3 times Y. and 75 |
|
|
|
181 |
|
00:15:47,210 --> 00:15:52,970 |
|
plus three times one. So that's 6 days in two |
|
|
|
182 |
|
00:15:52,970 --> 00:15:59,150 |
|
nights. Now let's look carefully at these three |
|
|
|
183 |
|
00:15:59,150 --> 00:16:04,910 |
|
intervals. The first one is seven to eight, the |
|
|
|
184 |
|
00:16:04,910 --> 00:16:11,050 |
|
other one 65 to 85, then 6 to 90. When we are |
|
|
|
185 |
|
00:16:11,050 --> 00:16:11,790 |
|
more confident, |
|
|
|
186 |
|
00:16:15,170 --> 00:16:20,630 |
|
When we are more confident here for 99.7%, the |
|
|
|
187 |
|
00:16:20,630 --> 00:16:25,930 |
|
interval becomes wider. So this is the widest |
|
|
|
188 |
|
00:16:25,930 --> 00:16:31,430 |
|
interval. Because here, the length of the interval |
|
|
|
189 |
|
00:16:31,430 --> 00:16:37,090 |
|
is around 10. The other one is 20. |
|
|
|
223 |
|
00:19:32,970 --> 00:19:36,370 |
|
falls within two standard ratios. That if the data |
|
|
|
224 |
|
00:19:36,370 --> 00:19:39,350 |
|
is bell shaped. Now what's about if the data is |
|
|
|
225 |
|
00:19:39,350 --> 00:19:43,210 |
|
not bell shaped? We have to use the shape-shape rule. |
|
|
|
226 |
|
00:19:43,830 --> 00:19:51,170 |
|
So 1 minus 1 over k is 2. So 2, 2, 2 squared. So 1 |
|
|
|
227 |
|
00:19:51,170 --> 00:19:58,130 |
|
minus 1 fourth. That gives. three quarters, I |
|
|
|
228 |
|
00:19:58,130 --> 00:20:03,370 |
|
mean, 75%. So instead of saying 95% of the data |
|
|
|
229 |
|
00:20:03,370 --> 00:20:06,850 |
|
lies within one or two standard deviations of the |
|
|
|
230 |
|
00:20:06,850 --> 00:20:13,070 |
|
mean, if the data is bell-shaped, if the data is |
|
|
|
231 |
|
00:20:13,070 --> 00:20:17,590 |
|
not bell-shaped, you have to say that 75% of the |
|
|
|
232 |
|
00:20:17,590 --> 00:20:22,190 |
|
data falls within two standard deviations. For |
|
|
|
233 |
|
00:20:22,190 --> 00:20:26,570 |
|
bell shape, you are 95% confident there. But here, |
|
|
|
234 |
|
00:20:27,190 --> 00:20:36,710 |
|
you're just 75% confident. Suppose k is 3. Now for |
|
|
|
235 |
|
00:20:36,710 --> 00:20:41,110 |
|
k equal 3, we said 99.7% of the data falls within |
|
|
|
236 |
|
00:20:41,110 --> 00:20:44,890 |
|
three standard deviations. Now here, if the data |
|
|
|
237 |
|
00:20:44,890 --> 00:20:51,940 |
|
is not bell shape, 1 minus 1 over k squared. 1 |
|
|
|
238 |
|
00:20:51,940 --> 00:20:56,540 |
|
minus 1 |
|
|
|
239 |
|
00:20:56,540 --> 00:21:00,760 |
|
over 3 squared is one-ninth. One-ninth is 0.11. 1 |
|
|
|
240 |
|
00:21:00,760 --> 00:21:06,440 |
|
minus 0.11 means 89% of the data, instead of |
|
|
|
241 |
|
00:21:06,440 --> 00:21:13,900 |
|
saying 99.7. So 89% of the data will fall within |
|
|
|
242 |
|
00:21:13,900 --> 00:21:16,460 |
|
three standard deviations of the population mean. |
|
|
|
243 |
|
00:21:18,510 --> 00:21:22,610 |
|
regardless of how the data are distributed around |
|
|
|
244 |
|
00:21:22,610 --> 00:21:26,350 |
|
them. So here, we have two scenarios. One, if the |
|
|
|
245 |
|
00:21:26,350 --> 00:21:29,390 |
|
data is symmetric, which is called the empirical rule |
|
|
|
246 |
|
00:21:29,390 --> 00:21:34,710 |
|
68959917. And the other one is called the shape-by |
|
|
|
247 |
|
00:21:34,710 --> 00:21:38,370 |
|
-shape rule, and that regardless of the shape of |
|
|
|
248 |
|
00:21:38,370 --> 00:21:38,710 |
|
the data. |
|
|
|
249 |
|
00:21:41,890 --> 00:21:49,210 |
|
Excuse me? Yes. In this case, you don't know the |
|
|
|
250 |
|
00:21:49,210 --> 00:21:51,490 |
|
distribution of the data. And the reality is |
|
|
|
251 |
|
00:21:51,490 --> 00:21:58,650 |
|
sometimes the data has an unknown distribution. For |
|
|
|
252 |
|
00:21:58,650 --> 00:22:02,590 |
|
this reason, we have to use chip-chip portions. |
|
|
|
253 |
|
00:22:05,410 --> 00:22:09,830 |
|
That's all for the empirical rule and the chip-chip rule. |
|
|
|
254 |
|
00:22:11,230 --> 00:22:18,150 |
|
The next topic is quartile measures. So far, we |
|
|
|
255 |
|
00:22:18,150 --> 00:22:24,330 |
|
have discussed central tendency measures, and we |
|
|
|
256 |
|
00:22:24,330 --> 00:22:28,450 |
|
have talked about mean, median, and more. Then we |
|
|
|
257 |
|
00:22:28,450 --> 00:22:32,830 |
|
moved to location of variability or spread or |
|
|
|
258 |
|
00:22:32,830 --> 00:22:37,810 |
|
dispersion. And we talked about range, variance, |
|
|
|
259 |
|
00:22:37,950 --> 00:22:38,890 |
|
and standardization. |
|
|
|
260 |
|
00:22:41,570 --> 00:22:48,230 |
|
And we said that outliers affect the mean much |
|
|
|
261 |
|
00:22:48,230 --> 00:22:51,470 |
|
more than the median. And also, outliers affect |
|
|
|
262 |
|
00:22:51,470 --> 00:22:55,730 |
|
the range. Here, we'll talk about other measures |
|
|
|
263 |
|
00:22:55,730 --> 00:22:59,570 |
|
of the data, which is called quartile measures. |
|
|
|
264 |
|
00:23:01,190 --> 00:23:03,450 |
|
Here, actually, we'll talk about two measures. |
|
|
|
265 |
|
00:23:04,270 --> 00:23:10,130 |
|
The first one is called the first quartile, and the other |
|
|
|
266 |
|
00:23:10,130 --> 00:23:14,150 |
|
one is the third quartile. So we have two measures, |
|
|
|
267 |
|
00:23:15,470 --> 00:23:26,030 |
|
the first and the third quartile. Quartiles split the ranked |
|
|
|
268 |
|
00:23:26,030 --> 00:23:32,930 |
|
data into four equal segments. I mean, these |
|
|
|
269 |
|
00:23:32,930 --> 00:23:37,190 |
|
measures split the data you have into four equal |
|
|
|
270 |
|
00:23:37,190 --> 00:23:37,730 |
|
parts. |
|
|
|
271 |
|
00:23:42,850 --> 00:23:48,690 |
|
Q1 has 25% of the data fall below it. I mean 25% |
|
|
|
272 |
|
00:23:48,690 --> 00:23:56,410 |
|
of the values lie below Q1. So it means 75% of the |
|
|
|
273 |
|
00:23:56,410 --> 00:24:04,410 |
|
values are above it. So 25 below and 75 above. But you |
|
|
|
274 |
|
00:24:04,410 --> 00:24:07,370 |
|
have to be careful that the data is arranged from |
|
|
|
275 |
|
00:24:07,370 --> 00:24:12,430 |
|
smallest to largest. So in this case, Q1 is a |
|
|
|
276 |
|
00:24:12,430 --> 00:24:19,630 |
|
value that has 25% below it. So Q2 is called the |
|
|
|
277 |
|
00:24:19,630 --> 00:24:22,450 |
|
median. The median, the value in the middle when |
|
|
|
278 |
|
00:24:22,450 --> 00:24:26,250 |
|
we arrange the data from smallest to largest. So |
|
|
|
279 |
|
00:24:26,250 --> 00:24:31,190 |
|
that means 50% of the data below and also 50% of |
|
|
|
280 |
|
00:24:31,190 --> 00:24:36,370 |
|
the data above. The other measure is called |
|
|
|
281 |
|
00:24:36,370 --> 00:24:41,730 |
|
the theoretical qualifying. In this case, we have 25% |
|
|
|
282 |
|
00:24:41,730 --> 00:24:47,950 |
|
of the data above Q3 and 75% of the data below Q3. |
|
|
|
283 |
|
00:24:49,010 --> 00:24:54,410 |
|
So quartiles split the ranked data into four equal |
|
|
|
284 |
|
00:24:54,410 --> 00:25:00,190 |
|
segments, Q1 25% to the left, Q2 50% to the left, |
|
|
|
285 |
|
00:25:00,970 --> 00:25:08,590 |
|
Q3 75% to the left, and 25% to the right. Before, |
|
|
|
286 |
|
00:25:09,190 --> 00:25:13,830 |
|
we explained how to compute the median, and let's |
|
|
|
287 |
|
00:25:13,830 --> 00:25:18,850 |
|
see how can we compute the first and third quartile. |
|
|
|
288 |
|
00:25:19,750 --> 00:25:23,650 |
|
If you remember, when we computed the median, |
|
|
|
289 |
|
00:25:24,350 --> 00:25:28,480 |
|
first we located the position of the median. And we |
|
|
|
290 |
|
00:25:28,480 --> 00:25:33,540 |
|
said that the rank of n is odd. Yes, it was n plus |
|
|
|
291 |
|
00:25:33,540 --> 00:25:37,800 |
|
1 divided by 2. This is the location of the |
|
|
|
292 |
|
00:25:37,800 --> 00:25:41,100 |
|
median, not the value. Sometimes the value may be |
|
|
|
293 |
|
00:25:41,100 --> 00:25:44,900 |
|
equal to the location, but most of the time it's |
|
|
|
294 |
|
00:25:44,900 --> 00:25:48,340 |
|
not. It's not the case. Now let's see how can we |
|
|
|
295 |
|
00:25:48,340 --> 00:25:54,130 |
|
locate the fair support. The first quartile after |
|
|
|
296 |
|
00:25:54,130 --> 00:25:56,690 |
|
you arrange the data from smallest to largest, the |
|
|
|
297 |
|
00:25:56,690 --> 00:26:01,290 |
|
location is n plus 1 divided by 2. So that's the |
|
|
|
298 |
|
00:26:01,290 --> 00:26:06,890 |
|
location of the first quartile. The median, as we |
|
|
|
299 |
|
00:26:06,890 --> 00:26:10,390 |
|
mentioned before, is located in the middle. So it |
|
|
|
300 |
|
00:26:10,390 --> 00:26:15,210 |
|
makes sense that if n is odd, the location of the |
|
|
|
301 |
|
00:26:15,210 --> 00:26:20,490 |
|
median is n plus 1 over 2. Now, for the third |
|
|
|
302 |
|
00:26:20,490 --> 00:26:27,160 |
|
quartile position, The location is N plus 1 |
|
|
|
303 |
|
00:26:27,160 --> 00:26:31,160 |
|
divided by 4 times 3. So 3 times N plus 1 divided |
|
|
|
304 |
|
00:26:31,160 --> 00:26:39,920 |
|
by 4. That's how can we locate Q1, Q2, and Q3. So |
|
|
|
305 |
|
00:26:39,920 --> 00:26:42,080 |
|
one more time, the median, the value in the |
|
|
|
306 |
|
00:26:42,080 --> 00:26:46,260 |
|
middle, and it's located exactly at the position N |
|
|
|
307 |
|
00:26:46,260 --> 00:26:52,590 |
|
plus 1 over 2 for the ranked data. Q1 is located at |
|
|
|
308 |
|
00:26:52,590 --> 00:26:56,770 |
|
n plus one divided by four. Q3 is located at the |
|
|
|
309 |
|
00:26:56,770 --> 00:26:59,670 |
|
position three times n plus one divided by four. |
|
|
|
310 |
|
00:27:03,630 --> 00:27:07,490 |
|
Now, when calculating the rank position, we can |
|
|
|
311 |
|
00:27:07,490 --> 00:27:14,690 |
|
use one of these rules. First, if the result of |
|
|
|
312 |
|
00:27:14,690 --> 00:27:18,010 |
|
the location, I mean, is a whole number, I mean, |
|
|
|
313 |
|
00:27:18,250 --> 00:27:24,050 |
|
if it is an integer. Then the rank position is the |
|
|
|
314 |
|
00:27:24,050 --> 00:27:28,590 |
|
same number. For example, suppose the rank |
|
|
|
315 |
|
00:27:28,590 --> 00:27:34,610 |
|
position is four. So position number four is your |
|
|
|
316 |
|
00:27:34,610 --> 00:27:38,450 |
|
quartile, either first or third or second |
|
|
|
317 |
|
00:27:38,450 --> 00:27:42,510 |
|
quartile. So if the result is a whole number, then |
|
|
|
318 |
|
00:27:42,510 --> 00:27:48,350 |
|
it is the rank position used. Now, if the result |
|
|
|
319 |
|
00:27:48,350 --> 00:27:52,250 |
|
is a fractional half, I mean if the right position |
|
|
|
320 |
|
00:27:52,250 --> 00:27:58,830 |
|
is 2.5, 3.5, 4.5. In this case, average the two |
|
|
|
321 |
|
00:27:58,830 --> 00:28:02,050 |
|
corresponding data values. For example, if the |
|
|
|
322 |
|
00:28:02,050 --> 00:28:10,170 |
|
right position is 2.5. So the rank position is 2 |
|
|
|
323 |
|
00:28:10,170 --> 00:28:13,210 |
|
.5. So take the average of the corresponding |
|
|
|
324 |
|
00:28:13,210 --> 00:28:18,950 |
|
values for the rank 2 and 3. So look at the value. |
|
|
|
325 |
|
00:28:19,280 --> 00:28:24,740 |
|
at rank 2, value at rank 3, then take the average |
|
|
|
326 |
|
00:28:24,740 --> 00:28:29,300 |
|
of the corresponding values. That if the rank |
|
|
|
327 |
|
00:28:29,300 --> 00:28:31,280 |
|
position is fractional. |
|
|
|
328 |
|
00:28:34,380 --> 00:28:37,900 |
|
So if the result is a whole number, just take it as |
|
|
|
329 |
|
00:28:37,900 --> 00:28:41,160 |
|
it is. If it is a fractional half, take the |
|
|
|
330 |
|
00:28:41,160 --> 00:28:44,460 |
|
corresponding data values and take the average of |
|
|
|
331 |
|
00:28:44,460 --> 00:28:49,110 |
|
these two values. Now, if the result is not a |
|
|
|
332 |
|
00:28:49,110 --> 00:28:53,930 |
|
whole number or a fraction of it. For example, |
|
|
|
333 |
|
00:28:54,070 --> 00:29:01,910 |
|
suppose the location is 2.1. So the position is 2, |
|
|
|
334 |
|
00:29:02,390 --> 00:29:06,550 |
|
just round up to the nearest integer. So that's |
|
|
|
335 |
|
00:29:06,550 --> 00:29:11,350 |
|
2. What's about if the position rank is 2.6? Just |
|
|
|
336 |
|
00:29:11,350 --> 00:29:16,060 |
|
rank up to 3. So that's 3. So that's the rule you |
|
|
|
337 |
|
00:29:16,060 --> 00:29:21,280 |
|
have to follow if the result is a number, a whole |
|
|
|
338 |
|
00:29:21,280 --> 00:29:27,200 |
|
number, I mean integer, fraction of half, or not |
|
|
|
339 |
|
00:29:27,200 --> 00:29:31,500 |
|
a real number, I mean, not whole number, or fraction |
|
|
|
340 |
|
00:29:31,500 --> 00:29:35,540 |
|
of half. Look at this specific example. Suppose we |
|
|
|
341 |
|
00:29:35,540 --> 00:29:40,180 |
|
have this data. This is an ordered array, 11, 12, up |
|
|
|
342 |
|
00:29:40,180 --> 00:29:45,680 |
|
to 22. And let's see how can we compute these |
|
|
|
343 |
|
00:29:45,680 --> 00:29:46,240 |
|
measures. |
|
|
|
344 |
|
00:29:50,080 --> 00:29:51,700 |
|
Look carefully here. |
|
|
|
345 |
|
00:29:55,400 --> 00:29:59,260 |
|
First, let's compute the median. The median is |
|
|
|
346 |
|
00:29:59,260 --> 00:30:02,360 |
|
the value in the middle. How many values do we have? |
|
|
|
347 |
|
00:30:02,800 --> 00:30:08,920 |
|
There are nine values. So the middle is number |
|
|
|
348 |
|
00:30:08,920 --> 00:30:15,390 |
|
five. One, two, three, four, five. So 16. This |
|
|
|
349 |
|
00:30:15,390 --> 00:30:23,010 |
|
value is the median. Now look at the values below |
|
|
|
350 |
|
00:30:23,010 --> 00:30:29,650 |
|
the median. There are four below and four above the |
|
|
|
351 |
|
00:30:29,650 --> 00:30:34,970 |
|
median. Now let's see how can we compute Q1. The |
|
|
|
352 |
|
00:30:34,970 --> 00:30:38,250 |
|
position of Q1, as we mentioned, is N plus 1 |
|
|
|
353 |
|
00:30:38,250 --> 00:30:42,630 |
|
divided by 4. So N is 9 plus 1 divided by 4 is 2 |
|
|
|
354 |
|
00:30:42,630 --> 00:30:50,330 |
|
.5. 2.5 position, it means you have to take the |
|
|
|
355 |
|
00:30:50,330 --> 00:30:54,490 |
|
average of the two corresponding values, 2 and 3. |
|
|
|
356 |
|
00:30:55,130 --> 00:31:01,010 |
|
So 2 and 3, so 12 plus 13 divided by 2. That gives |
|
|
|
357 |
|
00:31:01,010 --> 00:31:08,390 |
|
12.5. So this is Q1. |
|
|
|
358 |
|
00:31:08,530 --> 00:31:18,210 |
|
So Q1 is 12.5. Now what's about Q3? The Q3, the |
|
|
|
359 |
|
00:31:18,210 --> 00:31:27,810 |
|
rank position, Q1 was 2.5. So Q3 should be three |
|
|
|
360 |
|
00:31:27,810 --> 00:31:32,410 |
|
times that value, because it's three times A plus |
|
|
|
361 |
|
00:31:32,410 --> 00:31:36,090 |
|
1 over 4. That means the rank position is 7.5. |
|
|
|
362 |
|
00:31:36,590 --> 00:31:39,410 |
|
That means you have to take the average of the 7 |
|
|
|
363 |
|
00:31:39,410 --> 00:31:44,890 |
|
and 8 position. 7 and 8 is 18, |
|
|
|
364 |
|
00:31:45,880 --> 00:31:56,640 |
|
which is 19.5. So that's Q3, 19.5. |
|
|
|
365 |
|
00:32:00,360 --> 00:32:09,160 |
|
So this is Q3. This value is Q1. And this value |
|
|
|
366 |
|
00:32:09,160 --> 00:32:15,910 |
|
is? Now, Q2 is the center. It's located in the |
|
|
|
367 |
|
00:32:15,910 --> 00:32:18,570 |
|
center because, as we mentioned, four below and |
|
|
|
368 |
|
00:32:18,570 --> 00:32:22,950 |
|
four above. Now what's about Q1? Q1 is not in the |
|
|
|
369 |
|
00:32:22,950 --> 00:32:28,150 |
|
center of the entire data. Because Q1, 12.5, so |
|
|
|
370 |
|
00:32:28,150 --> 00:32:31,830 |
|
two points below and the others maybe how many |
|
|
|
371 |
|
00:32:31,830 --> 00:32:34,750 |
|
above, two, four, six, seven observations above it. |
|
|
|
372 |
|
00:32:35,390 --> 00:32:40,130 |
|
So that means Q1 is not the center. Also, Q3 is not |
|
|
|
373 |
|
00:32:40,130 --> 00:32:43,170 |
|
the center because two observations above it and seven |
|
|
|
374 |
|
00:32:43,170 --> 00:32:48,780 |
|
below it. So that means Q1 and Q3 are measures of |
|
|
|
375 |
|
00:32:48,780 --> 00:32:52,480 |
|
non-central location, while the median is a |
|
|
|
376 |
|
00:32:52,480 --> 00:32:56,080 |
|
measure of central location. But if you just look |
|
|
|
377 |
|
00:32:56,080 --> 00:33:03,720 |
|
at the data below the median, just focus on the |
|
|
|
378 |
|
00:33:03,720 --> 00:33:09,100 |
|
data below the median, 12.5 lies exactly in the |
|
|
|
379 |
|
00:33:09,100 --> 00:33:13,130 |
|
middle of the data. So 12.5 is the center of the |
|
|
|
380 |
|
00:33:13,130 --> 00:33:18,090 |
|
data. I mean, Q1 is the center of the data below |
|
|
|
381 |
|
00:33:18,090 --> 00:33:22,810 |
|
the overall median. The overall median was 16. So |
|
|
|
382 |
|
00:33:22,810 --> 00:33:27,490 |
|
the data before 16, the median for this data is 12 |
|
|
|
383 |
|
00:33:27,490 --> 00:33:31,770 |
|
.5, which is the first part. Similarly, if you |
|
|
|
384 |
|
00:33:31,770 --> 00:33:36,870 |
|
look at the data above Q2, |
|
|
|
385 |
|
00:33:37,770 --> 00:33:42,190 |
|
now 19.5 is located in the middle of the line. So |
|
|
|
386 |
|
00:33:42,190 --> 00:33:46,470 |
|
Q3 is a measure of the center for the data above the |
|
|
|
387 |
|
00:33:46,470 --> 00:33:48,390 |
|
line. Make sense? |
|
|
|
388 |
|
00:33:51,370 --> 00:33:56,430 |
|
So that's how we can compute the first, second, and |
|
|
|
389 |
|
00:33:56,430 --> 00:34:03,510 |
|
third part. Any questions? Yes, but it's a whole |
|
|
|
390 |
|
00:34:03,510 --> 00:34:09,370 |
|
number. Whole number, it means any integer. For |
|
|
|
391 |
|
00:34:09,370 --> 00:34:14,450 |
|
example, yeah, exactly, yes. Suppose we have |
|
|
|
392 |
|
00:34:14,450 --> 00:34:18,090 |
|
a number of data that is seven. |
|
|
|
393 |
|
00:34:22,070 --> 00:34:25,070 |
|
The number of observations we have is seven. So the |
|
|
|
394 |
|
00:34:25,070 --> 00:34:29,730 |
|
rank position, n plus one divided by two, seven |
|
|
|
395 |
|
00:34:29,730 --> 00:34:33,890 |
|
plus one over two is four. Four means a whole |
|
|
|
396 |
|
00:34:33,890 --> 00:34:37,780 |
|
number, I mean an integer. Then, in this case, just use |
|
|
|
397 |
|
00:34:37,780 --> 00:34:45,280 |
|
it as it is. Now let's see the benefit or the |
|
|
|
398 |
|
00:34:45,280 --> 00:34:48,680 |
|
feature of using Q1 and Q3. |
|
|
|
399 |
|
00:34:55,180 --> 00:35:01,300 |
|
So let's move on to the inter-quartile range or |
|
|
|
400 |
|
00:35:01,300 --> 00:35:01,760 |
|
IQR. |
|
|
|
401 |
|
00:35:08,020 --> 00:35:14,580 |
|
2.5 is the position. So the rank data of the ranked |
|
|
|
402 |
|
00:35:14,580 --> 00:35:19,180 |
|
data. So take the average of the two corresponding |
|
|
|
403 |
|
00:35:19,180 --> 00:35:25,700 |
|
values of this one, which are 2 and 3. So 2 and 3. |
|
|
|
404 |
|
00:35:27,400 --> |
|
|
|
445 |
|
00:39:11,650 --> 00:39:17,940 |
|
because it covers the middle 50% of the data. IQR |
|
|
|
446 |
|
00:39:17,940 --> 00:39:20,120 |
|
again is a measure of variability that is not |
|
|
|
447 |
|
00:39:20,120 --> 00:39:23,900 |
|
influenced or affected by outliers or extreme |
|
|
|
448 |
|
00:39:23,900 --> 00:39:26,680 |
|
values. So in the presence of outliers, it's |
|
|
|
449 |
|
00:39:26,680 --> 00:39:34,160 |
|
better to use IQR instead of using the range. So |
|
|
|
450 |
|
00:39:34,160 --> 00:39:39,140 |
|
again, median and the range are not affected by |
|
|
|
451 |
|
00:39:39,140 --> 00:39:43,180 |
|
outliers. So in case of the presence of outliers, |
|
|
|
452 |
|
00:39:43,340 --> 00:39:46,380 |
|
we have to use these measures, one as measure of |
|
|
|
453 |
|
00:39:46,380 --> 00:39:49,780 |
|
central and the other as measure of spread. So |
|
|
|
454 |
|
00:39:49,780 --> 00:39:54,420 |
|
measures like Q1, Q3, and IQR that are not |
|
|
|
455 |
|
00:39:54,420 --> 00:39:57,400 |
|
influenced by outliers are called resistant |
|
|
|
456 |
|
00:39:57,400 --> 00:40:01,980 |
|
measures. Resistance means in case of outliers, |
|
|
|
457 |
|
00:40:02,380 --> 00:40:06,120 |
|
they remain in the same position or approximately |
|
|
|
458 |
|
00:40:06,120 --> 00:40:09,870 |
|
in the same position. Because outliers don't |
|
|
|
459 |
|
00:40:09,870 --> 00:40:13,870 |
|
affect these measures. I mean, don't affect Q1, |
|
|
|
460 |
|
00:40:14,830 --> 00:40:20,130 |
|
Q3, and consequently IQR, because IQR is just the |
|
|
|
461 |
|
00:40:20,130 --> 00:40:24,990 |
|
distance between Q3 and Q1. So to determine the |
|
|
|
462 |
|
00:40:24,990 --> 00:40:29,430 |
|
value of IQR, you have first to compute Q1, Q3, |
|
|
|
463 |
|
00:40:29,750 --> 00:40:35,780 |
|
then take the difference between these two. So, |
|
|
|
464 |
|
00:40:36,120 --> 00:40:41,120 |
|
for example, suppose we have a data, and that data |
|
|
|
465 |
|
00:40:41,120 --> 00:40:51,400 |
|
has Q1 equals 30, and Q3 is 55. Suppose for a data |
|
|
|
466 |
|
00:40:51,400 --> 00:41:00,140 |
|
set, that data set has Q1 30, Q3 is 57. The IQR, |
|
|
|
467 |
|
00:41:00,800 --> 00:41:07,240 |
|
or Inter Equal Hyper Range, 57 minus 30 is 27. Now |
|
|
|
468 |
|
00:41:07,240 --> 00:41:12,460 |
|
what's the range? The range is maximum for the |
|
|
|
469 |
|
00:41:12,460 --> 00:41:17,380 |
|
largest value, which is 17 minus 12. That gives |
|
|
|
470 |
|
00:41:17,380 --> 00:41:21,420 |
|
58. Now look at the difference between the two |
|
|
|
471 |
|
00:41:21,420 --> 00:41:26,900 |
|
ranges. The inter-quartile range is 27. The range |
|
|
|
472 |
|
00:41:26,900 --> 00:41:29,800 |
|
is 58. There is a big difference between these two |
|
|
|
473 |
|
00:41:29,800 --> 00:41:35,750 |
|
values because range depends only on smallest and |
|
|
|
474 |
|
00:41:35,750 --> 00:41:40,190 |
|
largest. And these values could be outliers. For |
|
|
|
475 |
|
00:41:40,190 --> 00:41:44,410 |
|
this reason, the range value is higher or greater |
|
|
|
476 |
|
00:41:44,410 --> 00:41:48,410 |
|
than the required range, which is just the |
|
|
|
477 |
|
00:41:48,410 --> 00:41:54,050 |
|
distance of the 50% of the middle data. For this |
|
|
|
478 |
|
00:41:54,050 --> 00:41:59,470 |
|
reason, it's better to use the range in case of |
|
|
|
479 |
|
00:41:59,470 --> 00:42:03,940 |
|
outliers. Make sense? Any question? |
|
|
|
480 |
|
00:42:08,680 --> 00:42:19,320 |
|
Five-number summary are smallest |
|
|
|
481 |
|
00:42:19,320 --> 00:42:27,380 |
|
value, largest value, also first quartile, third |
|
|
|
482 |
|
00:42:27,380 --> 00:42:32,250 |
|
quartile, and the median. These five numbers are |
|
|
|
483 |
|
00:42:32,250 --> 00:42:35,870 |
|
called five-number summary, because by using these |
|
|
|
484 |
|
00:42:35,870 --> 00:42:41,590 |
|
statistics, smallest, first, median, third |
|
|
|
485 |
|
00:42:41,590 --> 00:42:46,010 |
|
quarter, and largest, you can describe the center |
|
|
|
486 |
|
00:42:46,010 --> 00:42:52,590 |
|
spread and the shape of the distribution. So by |
|
|
|
487 |
|
00:42:52,590 --> 00:42:56,450 |
|
using five-number summary, you can tell something |
|
|
|
488 |
|
00:42:56,450 --> 00:43:00,090 |
|
about it. The center of the data, I mean the value |
|
|
|
489 |
|
00:43:00,090 --> 00:43:02,070 |
|
in the middle, because the median is the value in |
|
|
|
490 |
|
00:43:02,070 --> 00:43:06,550 |
|
the middle. Spread, because we can talk about the |
|
|
|
491 |
|
00:43:06,550 --> 00:43:11,070 |
|
IQR, which is the range, and also the shape of the |
|
|
|
492 |
|
00:43:11,070 --> 00:43:15,450 |
|
data. And let's see, let's move to this slide, |
|
|
|
493 |
|
00:43:16,670 --> 00:43:18,530 |
|
slide number 50. |
|
|
|
494 |
|
00:43:21,530 --> 00:43:25,090 |
|
Let's see how can we construct something called |
|
|
|
495 |
|
00:43:25,090 --> 00:43:31,850 |
|
box plot. Box plot. Box plot can be constructed by |
|
|
|
496 |
|
00:43:31,850 --> 00:43:34,990 |
|
using the five number summary. We have smallest |
|
|
|
497 |
|
00:43:34,990 --> 00:43:37,550 |
|
value. On the other hand, we have the largest |
|
|
|
498 |
|
00:43:37,550 --> 00:43:43,430 |
|
value. Also, we have Q1, the first quartile, the |
|
|
|
499 |
|
00:43:43,430 --> 00:43:47,510 |
|
median, and Q3. For symmetric distribution, I mean |
|
|
|
500 |
|
00:43:47,510 --> 00:43:52,490 |
|
if the data is bell-shaped. In this case, the |
|
|
|
501 |
|
00:43:52,490 --> 00:43:56,570 |
|
vertical line in the box which represents the |
|
|
|
502 |
|
00:43:56,570 --> 00:43:59,730 |
|
median should be located in the middle of this |
|
|
|
503 |
|
00:43:59,730 --> 00:44:05,510 |
|
box, also in the middle of the entire data. Look |
|
|
|
504 |
|
00:44:05,510 --> 00:44:11,350 |
|
carefully at this vertical line. This line splits |
|
|
|
505 |
|
00:44:11,350 --> 00:44:16,070 |
|
the data into two halves, 25% to the left and 25% |
|
|
|
506 |
|
00:44:16,070 --> 00:44:19,960 |
|
to the right. And also this vertical line splits |
|
|
|
507 |
|
00:44:19,960 --> 00:44:24,720 |
|
the data into two halves, from the smallest to |
|
|
|
508 |
|
00:44:24,720 --> 00:44:29,760 |
|
largest, because there are 50% of the observations |
|
|
|
509 |
|
00:44:29,760 --> 00:44:34,560 |
|
lie below, and 50% lies above. So that means by |
|
|
|
510 |
|
00:44:34,560 --> 00:44:37,840 |
|
using box plot, you can tell something about the |
|
|
|
511 |
|
00:44:37,840 --> 00:44:42,520 |
|
shape of the distribution. So again, if the data |
|
|
|
512 |
|
00:44:42,520 --> 00:44:48,270 |
|
are symmetric around the median, And the central |
|
|
|
513 |
|
00:44:48,270 --> 00:44:53,910 |
|
line, this box, and central line are centered |
|
|
|
514 |
|
00:44:53,910 --> 00:44:57,550 |
|
between the endpoints. I mean, this vertical line |
|
|
|
515 |
|
00:44:57,550 --> 00:45:00,720 |
|
is centered between these two endpoints. between |
|
|
|
516 |
|
00:45:00,720 --> 00:45:04,180 |
|
Q1 and Q3. And the whole box plot is centered |
|
|
|
517 |
|
00:45:04,180 --> 00:45:07,100 |
|
between the smallest and the largest value. And |
|
|
|
518 |
|
00:45:07,100 --> 00:45:10,840 |
|
also the distance between the median and the |
|
|
|
519 |
|
00:45:10,840 --> 00:45:14,320 |
|
smallest is roughly equal to the distance between |
|
|
|
520 |
|
00:45:14,320 --> 00:45:19,760 |
|
the median and the largest. So you can tell |
|
|
|
521 |
|
00:45:19,760 --> 00:45:22,660 |
|
something about the shape of the distribution by |
|
|
|
522 |
|
00:45:22,660 --> 00:45:26,780 |
|
using the box plot. |
|
|
|
523 |
|
00:45:32,870 --> 00:45:36,110 |
|
The graph in the middle. Here median and median |
|
|
|
524 |
|
00:45:36,110 --> 00:45:40,110 |
|
are the same. The box plot, we have here the |
|
|
|
525 |
|
00:45:40,110 --> 00:45:43,830 |
|
median in the middle of the box, also in the |
|
|
|
526 |
|
00:45:43,830 --> 00:45:47,390 |
|
middle of the entire data. So you can say that the |
|
|
|
527 |
|
00:45:47,390 --> 00:45:50,210 |
|
distribution of this data is symmetric or is bell |
|
|
|
528 |
|
00:45:50,210 --> 00:45:55,750 |
|
-shaped. It's normal distribution. On the other |
|
|
|
529 |
|
00:45:55,750 --> 00:46:00,110 |
|
hand, if you look here, you will see that the |
|
|
|
530 |
|
00:46:00,110 --> 00:46:06,160 |
|
median is not in the center of the box. It's near |
|
|
|
531 |
|
00:46:06,160 --> 00:46:12,580 |
|
Q3. So the left tail, I mean, the distance between |
|
|
|
532 |
|
00:46:12,580 --> 00:46:16,620 |
|
the median and the smallest, this tail is longer |
|
|
|
533 |
|
00:46:16,620 --> 00:46:20,600 |
|
than the right tail. In this case, it's called |
|
|
|
534 |
|
00:46:20,600 --> 00:46:24,850 |
|
left skewed or skewed to the left. or negative |
|
|
|
535 |
|
00:46:24,850 --> 00:46:29,510 |
|
skewness. So if the data is not symmetric, it |
|
|
|
536 |
|
00:46:29,510 --> 00:46:35,630 |
|
might be left skewed. I mean, the left tail is |
|
|
|
537 |
|
00:46:35,630 --> 00:46:40,590 |
|
longer than the right tail. On the other hand, if |
|
|
|
538 |
|
00:46:40,590 --> 00:46:45,950 |
|
the median is located near Q1, it means the right |
|
|
|
539 |
|
00:46:45,950 --> 00:46:49,930 |
|
tail is longer than the left tail, and it's called |
|
|
|
540 |
|
00:46:49,930 --> 00:46:56,470 |
|
positive skewed or right skewed. So for symmetric |
|
|
|
541 |
|
00:46:56,470 --> 00:47:00,310 |
|
distribution, the median in the middle, for left |
|
|
|
542 |
|
00:47:00,310 --> 00:47:04,570 |
|
or right skewed, the median either is close to the |
|
|
|
543 |
|
00:47:04,570 --> 00:47:09,930 |
|
Q3 or skewed distribution to the left, or the |
|
|
|
544 |
|
00:47:09,930 --> 00:47:14,910 |
|
median is close to Q1 and the distribution is |
|
|
|
545 |
|
00:47:14,910 --> 00:47:20,570 |
|
right skewed or has positive skewness. That's how |
|
|
|
546 |
|
00:47:20,570 --> 00:47:25,860 |
|
can we tell spread center and the shape by using |
|
|
|
547 |
|
00:47:25,860 --> 00:47:28,460 |
|
the box plot. So center is the value in the |
|
|
|
548 |
|
00:47:28,460 --> 00:47:32,860 |
|
middle, Q2 or the median. Spread is the distance |
|
|
|
549 |
|
00:47:32,860 --> 00:47:38,340 |
|
between Q1 and Q3. So Q3 minus Q1 gives IQR. And |
|
|
|
550 |
|
00:47:38,340 --> 00:47:41,880 |
|
finally, you can tell something about the shape of |
|
|
|
551 |
|
00:47:41,880 --> 00:47:45,140 |
|
the distribution by just looking at the scatter |
|
|
|
552 |
|
00:47:45,140 --> 00:47:46,440 |
|
plot. |
|
|
|
553 |
|
00:47:49,700 --> 00:47:56,330 |
|
Let's look at This example, and suppose we have |
|
|
|
554 |
|
00:47:56,330 --> 00:48:02,430 |
|
small data set. And let's see how can we construct |
|
|
|
555 |
|
00:48:02,430 --> 00:48:05,750 |
|
the MaxPlot. In order to construct MaxPlot, you |
|
|
|
556 |
|
00:48:05,750 --> 00:48:09,510 |
|
have to compute minimum first or smallest value, |
|
|
|
557 |
|
00:48:09,810 --> 00:48:14,650 |
|
largest value. Besides that, you have to compute |
|
|
|
558 |
|
00:48:14,650 --> 00:48:21,110 |
|
first and third part time and also Q2. For this |
|
|
|
559 |
|
00:48:21,110 --> 00:48:27,570 |
|
simple example, Q1 is 2, Q3 is 5, and the median |
|
|
|
560 |
|
00:48:27,570 --> 00:48:33,990 |
|
is 3. Smallest is 0, largest is 17. Now, be |
|
|
|
561 |
|
00:48:33,990 --> 00:48:38,130 |
|
careful here, 17 seems to be an outlier. But so |
|
|
|
562 |
|
00:48:38,130 --> 00:48:44,190 |
|
far, we don't explain how can we decide if a data |
|
|
|
563 |
|
00:48:44,190 --> 00:48:47,550 |
|
value is considered to be an outlier. But at least |
|
|
|
564 |
|
00:48:47,550 --> 00:48:53,080 |
|
17. is a suspected value to be an outlier, seems |
|
|
|
565 |
|
00:48:53,080 --> 00:48:57,200 |
|
to be. Sometimes you are 95% sure that that point |
|
|
|
566 |
|
00:48:57,200 --> 00:49:00,160 |
|
is an outlier, but you cannot tell, because you |
|
|
|
567 |
|
00:49:00,160 --> 00:49:04,060 |
|
have to have a specific rule that can decide if |
|
|
|
568 |
|
00:49:04,060 --> 00:49:07,400 |
|
that point is an outlier or not. But at least it |
|
|
|
569 |
|
00:49:07,400 --> 00:49:12,060 |
|
makes sense that that point is considered maybe an |
|
|
|
570 |
|
00:49:12,060 --> 00:49:14,700 |
|
outlier. But let's see how can we construct that |
|
|
|
571 |
|
00:49:14,700 --> 00:49:18,190 |
|
first. The box plot. Again, as we mentioned, the |
|
|
|
572 |
|
00:49:18,190 --> 00:49:21,630 |
|
minimum value is zero. The maximum is 27. The Q1 |
|
|
|
573 |
|
00:49:21,630 --> 00:49:27,830 |
|
is 2. The median is 3. The Q3 is 5. Now, if you |
|
|
|
574 |
|
00:49:27,830 --> 00:49:32,010 |
|
look at the distance between, does this vertical |
|
|
|
575 |
|
00:49:32,010 --> 00:49:35,790 |
|
line lie between the line in the middle or the |
|
|
|
576 |
|
00:49:35,790 --> 00:49:40,090 |
|
center of the box? It's not exactly. But if you |
|
|
|
577 |
|
00:49:40,090 --> 00:49:45,260 |
|
look at this line, vertical line, and the location |
|
|
|
578 |
|
00:49:45,260 --> 00:49:50,600 |
|
of this with respect to the minimum and the |
|
|
|
579 |
|
00:49:50,600 --> 00:49:56,640 |
|
maximum. You will see that the right tail is much |
|
|
|
580 |
|
00:49:56,640 --> 00:50:01,560 |
|
longer than the left tail because it starts from 3 |
|
|
|
581 |
|
00:50:01,560 --> 00:50:06,180 |
|
up to 27. And the other one, from zero to three, |
|
|
|
582 |
|
00:50:06,380 --> 00:50:09,760 |
|
is a big distance between three and 27, compared |
|
|
|
583 |
|
00:50:09,760 --> 00:50:13,140 |
|
to the other one, zero to three. So it seems to be |
|
|
|
584 |
|
00:50:13,140 --> 00:50:16,600 |
|
this is quite skewed, so it's not at all |
|
|
|
585 |
|
00:50:16,600 --> 00:50:23,700 |
|
symmetric, because of this value. So maybe by |
|
|
|
586 |
|
00:50:23,700 --> 00:50:25,580 |
|
using MaxPlot, you can tell that point is |
|
|
|
587 |
|
00:50:25,580 --> 00:50:31,440 |
|
suspected to be an outlier. It has a very long |
|
|
|
588 |
|
00:50:31,440 --> 00:50:32,800 |
|
right tail. |
|
|
|
589 |
|
00:50:35,560 --> 00:50:41,120 |
|
So let's see how can we determine if a point is an |
|
|
|
590 |
|
00:50:41,120 --> 00:50:50,400 |
|
outlier or not. Sometimes we can use box plot to |
|
|
|
591 |
|
00:50:50,400 --> 00:50:53,840 |
|
determine if the point is an outlier or not. The |
|
|
|
592 |
|
00:50:53,840 --> 00:51:00,860 |
|
rule is that a value is considered an outlier It |
|
|
|
593 |
|
00:51:00,860 --> 00:51:04,780 |
|
is more than 1.5 times the entire quartile range |
|
|
|
594 |
|
00:51:04,780 --> 00:51:11,420 |
|
below Q1 or above it. Let's explain the meaning of |
|
|
|
595 |
|
00:51:11,420 --> 00:51:12,260 |
|
this sentence. |
|
|
|
596 |
|
00:51:15,260 --> 00:51:20,100 |
|
First, let's compute something called lower. |
|
|
|
597 |
|
00:51:23,740 --> 00:51:28,540 |
|
The lower limit is |
|
|
|
598 |
|
00:51:28,540 --> 00:51:38,680 |
|
not the minimum. It's Q1 minus 1.5 IQR. This is |
|
|
|
599 |
|
00:51:38,680 --> 00:51:39,280 |
|
the lower limit. |
|
|
|
600 |
|
00:51:42,280 --> 00:51:47,560 |
|
So it's 1.5 times IQR below Q1. This is the lower |
|
|
|
601 |
|
00:51:47,560 --> 00:51:50,620 |
|
limit. The upper limit, |
|
|
|
602 |
|
00:51:54,680 --> 00:51:57,460 |
|
Q3, |
|
|
|
603 |
|
00:51:58,790 --> 00:52:06,890 |
|
plus 1.5 times IQR. So we computed lower and upper |
|
|
|
604 |
|
00:52:06,890 --> 00:52:13,350 |
|
limit by using these rules. Q1 minus 1.5 IQR. So |
|
|
|
605 |
|
00:52:13,350 --> 00:52:20,510 |
|
it's 1.5 times IQR below Q1 and 1.5 times IQR |
|
|
|
606 |
|
00:52:20,510 --> 00:52:25,070 |
|
above Q1. Now, any value. |
|
|
|
607 |
|
00:52:31,150 --> 00:52:38,610 |
|
Is it smaller than the |
|
|
|
608 |
|
00:52:38,610 --> 00:52:45,990 |
|
lower limit or |
|
|
|
609 |
|
00:52:45,990 --> 00:52:53,290 |
|
greater than the |
|
|
|
610 |
|
00:52:53,290 --> 00:52:54,150 |
|
upper limit? |
|
|
|
611 |
|
00:52:58,330 --> 00:53:04,600 |
|
Any value. smaller than the lower limit and |
|
|
|
612 |
|
00:53:04,600 --> 00:53:13,260 |
|
greater than the upper limit is considered to |
|
|
|
613 |
|
00:53:13,260 --> 00:53:20,720 |
|
be an outlier. This is the rule how can you tell |
|
|
|
614 |
|
00:53:20,720 --> 00:53:24,780 |
|
if the point or data value is outlier or not. Just |
|
|
|
615 |
|
00:53:24,780 --> 00:53:27,100 |
|
compute lower limit and upper limit. |
|
|
|
616 |
|
00:53:29,780 --> 00:53:35,580 |
|
So lower limit, Q1 minus 1.5IQ3. Upper limit, Q3 |
|
|
|
617 |
|
00:53:35,580 --> 00:53:38,620 |
|
plus 1.5. This is a constant. |
|
|
|
618 |
|
00:53:43,200 --> 00:53:47,040 |
|
Now let's go back to the previous example, which |
|
|
|
619 |
|
00:53:47,040 --> 00:53:53,800 |
|
was, which Q1 was, what's the value of Q1? Q1 was |
|
|
|
620 |
|
00:53:53,800 --> 00:53:57,680 |
|
2. Q3 is 5. |
|
|
|
621 |
|
00:54:00,650 --> 00:54:05,230 |
|
In order to turn an outlier, you don't need the |
|
|
|
622 |
|
00:54:05,230 --> 00:54:11,150 |
|
value, the median. Now, Q3 is 5, Q1 is 2, so IQR |
|
|
|
623 |
|
00:54:11,150 --> 00:54:21,050 |
|
is 3. That's the value of IQR. Now, lower limit, A |
|
|
|
624 |
|
00:54:21,050 --> 00:54:31,830 |
|
times 2 minus 1.5 times IQR3. So that's minus 2.5. |
|
|
|
625 |
|
00:54:33,550 --> 00:54:41,170 |
|
U3 plus U3 is 3. It's 5, sorry. It's 5 plus 1.5. |
|
|
|
626 |
|
00:54:41,650 --> 00:54:48,570 |
|
That gives 9.5. Now, any point or any data value, |
|
|
|
627 |
|
00:54:49,450 --> 00:54:55,950 |
|
any data value falls below minus 2.5. I mean |
|
|
|
628 |
|
00:54:55,950 --> 00:55:00,380 |
|
smaller than minus 2. |