|
|
|
1 |
|
00:00:11,020 --> 00:00:13,920 |
|
The last chapter we are going to talk about in this |
|
|
|
2 |
|
00:00:13,920 --> 00:00:17,820 |
|
semester is correlation and simple linear regression. |
|
|
|
3 |
|
00:00:18,380 --> 00:00:23,300 |
|
So we are going to explain two types in chapter |
|
|
|
4 |
|
00:00:23,300 --> 00:00:29,280 |
|
12. One is called correlation. And the other type |
|
|
|
5 |
|
00:00:29,280 --> 00:00:33,500 |
|
is simple linear regression. Maybe this chapter |
|
|
|
6 |
|
00:00:33,500 --> 00:00:40,020 |
|
I'm going to spend about two lectures in order to |
|
|
|
7 |
|
00:00:40,020 --> 00:00:45,000 |
|
cover these objectives. The first objective is to |
|
|
|
8 |
|
00:00:45,000 --> 00:00:48,810 |
|
calculate the coefficient of correlation. The |
|
|
|
9 |
|
00:00:48,810 --> 00:00:51,210 |
|
second objective, the meaning of the regression |
|
|
|
10 |
|
00:00:51,210 --> 00:00:55,590 |
|
coefficients beta 0 and beta 1. And the last |
|
|
|
11 |
|
00:00:55,590 --> 00:00:58,710 |
|
objective is how to use regression analysis to |
|
|
|
12 |
|
00:00:58,710 --> 00:01:03,030 |
|
predict the value of dependent variable based on |
|
|
|
13 |
|
00:01:03,030 --> 00:01:06,010 |
|
an independent variable. It looks like that we |
|
|
|
14 |
|
00:01:06,010 --> 00:01:10,590 |
|
have discussed objective number one in chapter |
|
|
|
15 |
|
00:01:10,590 --> 00:01:16,470 |
|
three. So calculation of the correlation |
|
|
|
16 |
|
00:01:16,470 --> 00:01:20,740 |
|
coefficient is done in chapter three, but here |
|
|
|
17 |
|
00:01:20,740 --> 00:01:26,060 |
|
we'll give some details about correlation also. A |
|
|
|
18 |
|
00:01:26,060 --> 00:01:28,480 |
|
scatter plot can be used to show the relationship |
|
|
|
19 |
|
00:01:28,480 --> 00:01:31,540 |
|
between two variables. For example, imagine that |
|
|
|
20 |
|
00:01:31,540 --> 00:01:35,400 |
|
we have a random sample of 10 children. |
|
|
|
21 |
|
00:01:37,800 --> 00:01:47,940 |
|
And we have data on their weights and ages. And we |
|
|
|
22 |
|
00:01:47,940 --> 00:01:51,640 |
|
are interested to examine the relationship between |
|
|
|
23 |
|
00:01:51,640 --> 00:01:58,400 |
|
weights and age. For example, suppose child number |
|
|
|
24 |
|
00:01:58,400 --> 00:02:06,260 |
|
one, his |
|
|
|
25 |
|
00:02:06,260 --> 00:02:12,060 |
|
or her age is two years with weight, for example, |
|
|
|
26 |
|
00:02:12,200 --> 00:02:12,880 |
|
eight kilograms. |
|
|
|
27 |
|
00:02:17,680 --> 00:02:21,880 |
|
His weight or her weight is four years, and his or |
|
|
|
28 |
|
00:02:21,880 --> 00:02:24,500 |
|
her weight is, for example, 15 kilograms, and so |
|
|
|
29 |
|
00:02:24,500 --> 00:02:29,680 |
|
on. And again, we are interested to examine the |
|
|
|
30 |
|
00:02:29,680 --> 00:02:32,640 |
|
relationship between age and weight. Maybe they |
|
|
|
31 |
|
00:02:32,640 --> 00:02:37,400 |
|
exist sometimes. positive relationship between the |
|
|
|
32 |
|
00:02:37,400 --> 00:02:41,100 |
|
two variables that means if one variable increases |
|
|
|
33 |
|
00:02:41,100 --> 00:02:45,260 |
|
the other one also increase if one variable |
|
|
|
34 |
|
00:02:45,260 --> 00:02:47,980 |
|
increases the other will also decrease so they |
|
|
|
35 |
|
00:02:47,980 --> 00:02:52,980 |
|
have the same direction either up or down so we |
|
|
|
36 |
|
00:02:52,980 --> 00:02:58,140 |
|
have to know number one the form of the |
|
|
|
37 |
|
00:02:58,140 --> 00:03:02,140 |
|
relationship this one could be linear here we |
|
|
|
38 |
|
00:03:02,140 --> 00:03:06,890 |
|
focus just on linear relationship between X and Y. |
|
|
|
39 |
|
00:03:08,050 --> 00:03:13,730 |
|
The second, we have to know the direction of the |
|
|
|
40 |
|
00:03:13,730 --> 00:03:21,270 |
|
relationship. This direction might be positive or |
|
|
|
41 |
|
00:03:21,270 --> 00:03:22,350 |
|
negative relationship. |
|
|
|
42 |
|
00:03:25,150 --> 00:03:27,990 |
|
In addition to that, we have to know the strength |
|
|
|
43 |
|
00:03:27,990 --> 00:03:33,760 |
|
of the relationship between the two variables of |
|
|
|
44 |
|
00:03:33,760 --> 00:03:37,320 |
|
interest the strength can be classified into three |
|
|
|
45 |
|
00:03:37,320 --> 00:03:46,480 |
|
categories either strong, moderate or there exists |
|
|
|
46 |
|
00:03:46,480 --> 00:03:50,580 |
|
a weak relationship so it could be positive |
|
|
|
47 |
|
00:03:50,580 --> 00:03:53,320 |
|
-strong, positive-moderate or positive-weak, the |
|
|
|
48 |
|
00:03:53,320 --> 00:03:58,360 |
|
same for negative so by using scatter plot we can |
|
|
|
49 |
|
00:03:58,360 --> 00:04:02,530 |
|
determine the form either linear or non-linear, |
|
|
|
50 |
|
00:04:02,690 --> 00:04:06,130 |
|
but here we are focusing on just linear |
|
|
|
51 |
|
00:04:06,130 --> 00:04:10,310 |
|
relationship. Also, we can determine the direction |
|
|
|
52 |
|
00:04:10,310 --> 00:04:12,870 |
|
of the relationship. We can say there exists |
|
|
|
53 |
|
00:04:12,870 --> 00:04:15,910 |
|
positive or negative based on the scatter plot. |
|
|
|
54 |
|
00:04:16,710 --> 00:04:19,530 |
|
Also, we can know the strength of the |
|
|
|
55 |
|
00:04:19,530 --> 00:04:23,130 |
|
relationship, either strong, moderate or weak. For |
|
|
|
56 |
|
00:04:23,130 --> 00:04:29,810 |
|
example, suppose we have again weights and ages. |
|
|
|
57 |
|
00:04:30,390 --> 00:04:33,590 |
|
And we know that there are two types of variables |
|
|
|
58 |
|
00:04:33,590 --> 00:04:36,710 |
|
in this case. One is called dependent and the |
|
|
|
59 |
|
00:04:36,710 --> 00:04:41,330 |
|
other is independent. So if we, as we explained |
|
|
|
60 |
|
00:04:41,330 --> 00:04:47,890 |
|
before, is the dependent variable and A is |
|
|
|
61 |
|
00:04:47,890 --> 00:04:48,710 |
|
independent variable. |
|
|
|
62 |
|
00:04:52,690 --> 00:04:57,270 |
|
Always dependent |
|
|
|
63 |
|
00:04:57,270 --> 00:04:57,750 |
|
variable |
|
|
|
64 |
|
00:05:00,400 --> 00:05:05,560 |
|
is denoted by Y and always on the vertical axis so |
|
|
|
65 |
|
00:05:05,560 --> 00:05:11,300 |
|
here we have weight and independent variable is |
|
|
|
66 |
|
00:05:11,300 --> 00:05:17,760 |
|
denoted by X and X is in the X axis or horizontal |
|
|
|
67 |
|
00:05:17,760 --> 00:05:26,300 |
|
axis now scatter plot for example here child with |
|
|
|
68 |
|
00:05:26,300 --> 00:05:30,820 |
|
age 2 years his weight is 8 So two years, for |
|
|
|
69 |
|
00:05:30,820 --> 00:05:36,760 |
|
example, this is eight. So this star represents |
|
|
|
70 |
|
00:05:36,760 --> 00:05:42,320 |
|
the first pair of observation, age of two and |
|
|
|
71 |
|
00:05:42,320 --> 00:05:46,820 |
|
weight of eight. The other child, his weight is |
|
|
|
72 |
|
00:05:46,820 --> 00:05:52,860 |
|
four years, and the corresponding weight is 15. |
|
|
|
73 |
|
00:05:53,700 --> 00:05:58,970 |
|
For example, this value is 15. The same for the |
|
|
|
74 |
|
00:05:58,970 --> 00:06:02,430 |
|
other points. Here we can know the direction. |
|
|
|
75 |
|
00:06:04,910 --> 00:06:10,060 |
|
In this case they exist. Positive. Form is linear. |
|
|
|
76 |
|
00:06:12,100 --> 00:06:16,860 |
|
Strong or weak or moderate depends on how these |
|
|
|
77 |
|
00:06:16,860 --> 00:06:20,260 |
|
values are close to the straight line. Closer |
|
|
|
78 |
|
00:06:20,260 --> 00:06:24,380 |
|
means stronger. So if the points are closer to the |
|
|
|
79 |
|
00:06:24,380 --> 00:06:26,620 |
|
straight line, it means there exists stronger |
|
|
|
80 |
|
00:06:26,620 --> 00:06:30,800 |
|
relationship between the two variables. So closer |
|
|
|
81 |
|
00:06:30,800 --> 00:06:34,480 |
|
means stronger, either positive or negative. In |
|
|
|
82 |
|
00:06:34,480 --> 00:06:37,580 |
|
this case, there exists positive. Now for the |
|
|
|
83 |
|
00:06:37,580 --> 00:06:42,360 |
|
negative association or relationship, we have the |
|
|
|
84 |
|
00:06:42,360 --> 00:06:46,060 |
|
other direction, it could be this one. So in this |
|
|
|
85 |
|
00:06:46,060 --> 00:06:49,460 |
|
case there exists linear but negative |
|
|
|
86 |
|
00:06:49,460 --> 00:06:51,900 |
|
relationship, and this negative could be positive |
|
|
|
87 |
|
00:06:51,900 --> 00:06:56,100 |
|
or negative, it depends on the points. So it's |
|
|
|
88 |
|
00:06:56,100 --> 00:07:02,660 |
|
positive relationship. The other direction is |
|
|
|
89 |
|
00:07:02,660 --> 00:07:06,460 |
|
negative. So the points, if the points are closed, |
|
|
|
90 |
|
00:07:06,820 --> 00:07:10,160 |
|
then we can say there exists strong negative |
|
|
|
91 |
|
00:07:10,160 --> 00:07:14,440 |
|
relationship. So by using scatter plot, we can |
|
|
|
92 |
|
00:07:14,440 --> 00:07:17,280 |
|
determine all of these. |
|
|
|
93 |
|
00:07:20,840 --> 00:07:24,460 |
|
and direction and strength now here the two |
|
|
|
94 |
|
00:07:24,460 --> 00:07:27,060 |
|
variables we are talking about are numerical |
|
|
|
95 |
|
00:07:27,060 --> 00:07:30,480 |
|
variables so the two variables here are numerical |
|
|
|
96 |
|
00:07:30,480 --> 00:07:35,220 |
|
variables so we are talking about quantitative |
|
|
|
97 |
|
00:07:35,220 --> 00:07:39,850 |
|
variables but remember in chapter 11 We talked |
|
|
|
98 |
|
00:07:39,850 --> 00:07:43,150 |
|
about the relationship between two qualitative |
|
|
|
99 |
|
00:07:43,150 --> 00:07:47,450 |
|
variables. So we use chi-square test. Here we are |
|
|
|
100 |
|
00:07:47,450 --> 00:07:49,630 |
|
talking about something different. We are talking |
|
|
|
101 |
|
00:07:49,630 --> 00:07:52,890 |
|
about numerical variables. So we can use scatter |
|
|
|
102 |
|
00:07:52,890 --> 00:07:58,510 |
|
plot, number one. Next correlation analysis is |
|
|
|
103 |
|
00:07:58,510 --> 00:08:02,090 |
|
used to measure the strength of the association |
|
|
|
104 |
|
00:08:02,090 --> 00:08:05,190 |
|
between two variables. And here again, we are just |
|
|
|
105 |
|
00:08:05,190 --> 00:08:09,560 |
|
talking about linear relationship. So this chapter |
|
|
|
106 |
|
00:08:09,560 --> 00:08:13,340 |
|
just covers the linear relationship between the |
|
|
|
107 |
|
00:08:13,340 --> 00:08:17,040 |
|
two variables. Because sometimes there exists non |
|
|
|
108 |
|
00:08:17,040 --> 00:08:23,180 |
|
-linear relationship between the two variables. So |
|
|
|
109 |
|
00:08:23,180 --> 00:08:26,120 |
|
correlation is only concerned with the strength of |
|
|
|
110 |
|
00:08:26,120 --> 00:08:30,500 |
|
the relationship. No causal effect is implied with |
|
|
|
111 |
|
00:08:30,500 --> 00:08:35,220 |
|
correlation. We just say that X affects Y, or X |
|
|
|
112 |
|
00:08:35,220 --> 00:08:39,580 |
|
explains the variation in Y. Scatter plots were |
|
|
|
113 |
|
00:08:39,580 --> 00:08:43,720 |
|
first presented in Chapter 2, and we skipped, if |
|
|
|
114 |
|
00:08:43,720 --> 00:08:48,480 |
|
you remember, Chapter 2. And it's easy to make |
|
|
|
115 |
|
00:08:48,480 --> 00:08:52,620 |
|
scatter plots for Y versus X. In Chapter 3, we |
|
|
|
116 |
|
00:08:52,620 --> 00:08:56,440 |
|
talked about correlation, so correlation was first |
|
|
|
117 |
|
00:08:56,440 --> 00:09:00,060 |
|
presented in Chapter 3. But here I will give just |
|
|
|
118 |
|
00:09:00,060 --> 00:09:07,240 |
|
a review for computation about correlation |
|
|
|
119 |
|
00:09:07,240 --> 00:09:11,460 |
|
coefficient or coefficient of correlation. First, |
|
|
|
120 |
|
00:09:12,800 --> 00:09:15,680 |
|
coefficient of correlation measures the relative |
|
|
|
121 |
|
00:09:15,680 --> 00:09:19,920 |
|
strength of the linear relationship between two |
|
|
|
122 |
|
00:09:19,920 --> 00:09:23,740 |
|
numerical variables. So here, we are talking about |
|
|
|
123 |
|
00:09:23,740 --> 00:09:28,080 |
|
numerical variables. Sample correlation |
|
|
|
124 |
|
00:09:28,080 --> 00:09:31,500 |
|
coefficient is given by this equation. which is |
|
|
|
125 |
|
00:09:31,500 --> 00:09:36,180 |
|
sum of the product of xi minus x bar, yi minus y |
|
|
|
126 |
|
00:09:36,180 --> 00:09:41,100 |
|
bar, divided by n minus 1 times standard deviation |
|
|
|
127 |
|
00:09:41,100 --> 00:09:44,960 |
|
of x times standard deviation of y. We know that x |
|
|
|
128 |
|
00:09:44,960 --> 00:09:47,240 |
|
bar and y bar are the means of x and y |
|
|
|
129 |
|
00:09:47,240 --> 00:09:51,360 |
|
respectively. And Sx, Sy are the standard |
|
|
|
130 |
|
00:09:51,360 --> 00:09:55,540 |
|
deviations of x and y values. And we know this |
|
|
|
131 |
|
00:09:55,540 --> 00:09:58,460 |
|
equation before. But there is another equation |
|
|
|
132 |
|
00:09:58,460 --> 00:10:05,330 |
|
that one can be used For computation, which is |
|
|
|
133 |
|
00:10:05,330 --> 00:10:09,290 |
|
called shortcut formula, which is just sum of xy |
|
|
|
134 |
|
00:10:09,290 --> 00:10:15,310 |
|
minus n times x bar y bar divided by square root |
|
|
|
135 |
|
00:10:15,310 --> 00:10:18,690 |
|
of this quantity. And we know this equation from |
|
|
|
136 |
|
00:10:18,690 --> 00:10:23,650 |
|
chapter three. Now again, x bar and y bar are the |
|
|
|
137 |
|
00:10:23,650 --> 00:10:30,060 |
|
means. Now the question is, Do outliers affect the |
|
|
|
138 |
|
00:10:30,060 --> 00:10:36,440 |
|
correlation? For sure, yes. Because this formula |
|
|
|
139 |
|
00:10:36,440 --> 00:10:39,940 |
|
actually based on the means and the standard |
|
|
|
140 |
|
00:10:39,940 --> 00:10:44,300 |
|
deviations, and these two measures are affected by |
|
|
|
141 |
|
00:10:44,300 --> 00:10:47,880 |
|
outliers. So since R is a function of these two |
|
|
|
142 |
|
00:10:47,880 --> 00:10:51,340 |
|
statistics, the means and standard deviations, |
|
|
|
143 |
|
00:10:51,940 --> 00:10:54,280 |
|
then outliers will affect the value of the |
|
|
|
144 |
|
00:10:54,280 --> 00:10:55,940 |
|
correlation coefficient. |
|
|
|
145 |
|
00:10:57,890 --> 00:11:01,170 |
|
Some features about the coefficient of |
|
|
|
146 |
|
00:11:01,170 --> 00:11:09,570 |
|
correlation. Here rho is the population |
|
|
|
147 |
|
00:11:09,570 --> 00:11:13,210 |
|
coefficient of correlation, and R is the sample |
|
|
|
148 |
|
00:11:13,210 --> 00:11:17,730 |
|
coefficient of correlation. Either rho or R have |
|
|
|
149 |
|
00:11:17,730 --> 00:11:21,390 |
|
the following features. Number one, unity free. It |
|
|
|
150 |
|
00:11:21,390 --> 00:11:24,890 |
|
means R has no units. For example, here we are |
|
|
|
151 |
|
00:11:24,890 --> 00:11:28,820 |
|
talking about whales. And weight in kilograms, |
|
|
|
152 |
|
00:11:29,300 --> 00:11:33,700 |
|
ages in years. And for example, suppose the |
|
|
|
153 |
|
00:11:33,700 --> 00:11:37,080 |
|
correlation between these two variables is 0.8. |
|
|
|
154 |
|
00:11:38,620 --> 00:11:41,760 |
|
It's unity free, so it's just 0.8. So there is no |
|
|
|
155 |
|
00:11:41,760 --> 00:11:45,640 |
|
unit. You cannot say 0.8 kilogram per year or |
|
|
|
156 |
|
00:11:45,640 --> 00:11:51,040 |
|
whatever it is. So just 0.8. So the first feature |
|
|
|
157 |
|
00:11:51,040 --> 00:11:53,360 |
|
of the correlation coefficient is unity-free. |
|
|
|
158 |
|
00:11:54,180 --> 00:11:56,340 |
|
Number two ranges between negative one and plus |
|
|
|
159 |
|
00:11:56,340 --> 00:12:00,380 |
|
one. So R is always, or rho, is always between |
|
|
|
160 |
|
00:12:00,380 --> 00:12:04,560 |
|
minus one and plus one. So minus one smaller than |
|
|
|
161 |
|
00:12:04,560 --> 00:12:07,340 |
|
or equal to R smaller than or equal to plus one. |
|
|
|
162 |
|
00:12:07,420 --> 00:12:11,420 |
|
So R is always in this range. So R cannot be |
|
|
|
163 |
|
00:12:11,420 --> 00:12:15,260 |
|
smaller than negative one or greater than plus |
|
|
|
164 |
|
00:12:15,260 --> 00:12:20,310 |
|
one. The closer to minus one or negative one, the |
|
|
|
165 |
|
00:12:20,310 --> 00:12:23,130 |
|
stronger negative relationship between or linear |
|
|
|
166 |
|
00:12:23,130 --> 00:12:26,770 |
|
relationship between x and y. So, for example, if |
|
|
|
167 |
|
00:12:26,770 --> 00:12:33,370 |
|
R is negative 0.85 or R is negative 0.8. Now, this |
|
|
|
168 |
|
00:12:33,370 --> 00:12:39,690 |
|
value is closer to minus one than negative 0.8. So |
|
|
|
169 |
|
00:12:39,690 --> 00:12:43,230 |
|
negative 0.85 is stronger than negative 0.8. |
|
|
|
170 |
|
00:12:44,590 --> 00:12:48,470 |
|
Because we are looking for closer to minus 1. |
|
|
|
171 |
|
00:12:49,570 --> 00:12:55,310 |
|
Minus 0.8, the value itself is greater than minus |
|
|
|
172 |
|
00:12:55,310 --> 00:12:59,610 |
|
0.85. But this value is closer to minus 1 than |
|
|
|
173 |
|
00:12:59,610 --> 00:13:03,790 |
|
minus 0.8. So we can say that this relationship is |
|
|
|
174 |
|
00:13:03,790 --> 00:13:05,070 |
|
stronger than the other one. |
|
|
|
175 |
|
00:13:07,870 --> 00:13:11,730 |
|
Also, the closer to plus 1, the stronger the |
|
|
|
176 |
|
00:13:11,730 --> 00:13:16,040 |
|
positive linear relationship. Here, suppose R is 0 |
|
|
|
177 |
|
00:13:16,040 --> 00:13:22,740 |
|
.7 and another R is 0.8. 0.8 is closer to plus one |
|
|
|
178 |
|
00:13:22,740 --> 00:13:26,740 |
|
than 0.7, so 0.8 is stronger. This one makes |
|
|
|
179 |
|
00:13:26,740 --> 00:13:31,800 |
|
sense. The closer to zero, the weaker relationship |
|
|
|
180 |
|
00:13:31,800 --> 00:13:35,420 |
|
between the two variables. For example, suppose R |
|
|
|
181 |
|
00:13:35,420 --> 00:13:40,720 |
|
is plus or minus 0.05. This value is very close to |
|
|
|
182 |
|
00:13:40,720 --> 00:13:44,420 |
|
zero. It means there exists weak. relationship. |
|
|
|
183 |
|
00:13:44,980 --> 00:13:47,960 |
|
Sometimes we can say that there exists moderate |
|
|
|
184 |
|
00:13:47,960 --> 00:13:57,080 |
|
relationship if R is close to 0.5. So it could be |
|
|
|
185 |
|
00:13:57,080 --> 00:14:01,360 |
|
classified into these groups closer to minus 1, |
|
|
|
186 |
|
00:14:01,500 --> 00:14:06,220 |
|
closer to 1, 0.5 or 0. So we can know the |
|
|
|
187 |
|
00:14:06,220 --> 00:14:11,680 |
|
direction by the sign of R negative it means |
|
|
|
188 |
|
00:14:11,680 --> 00:14:14,320 |
|
because here our ranges as we mentioned between |
|
|
|
189 |
|
00:14:14,320 --> 00:14:19,520 |
|
minus one and plus one here zero so this these |
|
|
|
190 |
|
00:14:19,520 --> 00:14:24,560 |
|
values it means there exists negative above zero |
|
|
|
191 |
|
00:14:24,560 --> 00:14:26,760 |
|
all the way up to one it means there exists |
|
|
|
192 |
|
00:14:26,760 --> 00:14:31,020 |
|
positive relationship between the two variables so |
|
|
|
193 |
|
00:14:31,020 --> 00:14:35,520 |
|
the sign gives the direction of the relationship |
|
|
|
194 |
|
00:14:36,720 --> 00:14:40,840 |
|
The absolute value gives the strength of the |
|
|
|
195 |
|
00:14:40,840 --> 00:14:43,500 |
|
relationship between the two variables. So the |
|
|
|
196 |
|
00:14:43,500 --> 00:14:49,260 |
|
same as we had discussed before. Now, some types |
|
|
|
197 |
|
00:14:49,260 --> 00:14:51,880 |
|
of scatter plots for different types of |
|
|
|
198 |
|
00:14:51,880 --> 00:14:54,740 |
|
relationship between the two variables is |
|
|
|
199 |
|
00:14:54,740 --> 00:14:59,100 |
|
presented in this |
|
|
|
223 |
|
00:16:53,260 --> 00:16:58,860 |
|
individual and also weights are increased by three |
|
|
|
224 |
|
00:16:58,860 --> 00:17:03,080 |
|
units for each person. In this case there exists |
|
|
|
225 |
|
00:17:03,080 --> 00:17:06,820 |
|
a perfect relationship but that never happened in |
|
|
|
226 |
|
00:17:06,820 --> 00:17:13,300 |
|
real life. So perfect means all points are lie on |
|
|
|
227 |
|
00:17:13,300 --> 00:17:16,260 |
|
the straight line otherwise if the points are |
|
|
|
228 |
|
00:17:16,260 --> 00:17:21,230 |
|
close Then we can say there exists strong. Here if |
|
|
|
229 |
|
00:17:21,230 --> 00:17:24,750 |
|
you look carefully at these points corresponding |
|
|
|
230 |
|
00:17:24,750 --> 00:17:30,150 |
|
to this regression line, it looks like not strong |
|
|
|
231 |
|
00:17:30,150 --> 00:17:32,630 |
|
because some of the points are not close, so you |
|
|
|
232 |
|
00:17:32,630 --> 00:17:35,450 |
|
can say there exists maybe moderate negative |
|
|
|
233 |
|
00:17:35,450 --> 00:17:39,530 |
|
relationship. This one, most of the points are |
|
|
|
234 |
|
00:17:39,530 --> 00:17:42,390 |
|
scattered away from the straight line, so there |
|
|
|
235 |
|
00:17:42,390 --> 00:17:46,930 |
|
exists weak relationship. So by just looking at |
|
|
|
236 |
|
00:17:46,930 --> 00:17:50,290 |
|
the scatter path, sometimes you can, sometimes |
|
|
|
237 |
|
00:17:50,290 --> 00:17:53,290 |
|
it's hard to tell, but most of the time you can |
|
|
|
238 |
|
00:17:53,290 --> 00:17:58,250 |
|
tell at least the direction, positive or negative, |
|
|
|
239 |
|
00:17:59,410 --> 00:18:04,150 |
|
the form, linear or non-linear, or the strength of |
|
|
|
240 |
|
00:18:04,150 --> 00:18:09,100 |
|
the relationship. The last one here, now x |
|
|
|
241 |
|
00:18:09,100 --> 00:18:13,800 |
|
increases, y remains the same. For example, |
|
|
|
242 |
|
00:18:13,880 --> 00:18:18,580 |
|
suppose x is 1, y is 10. x increases to 2, y still |
|
|
|
243 |
|
00:18:18,580 --> 00:18:22,220 |
|
is 10. So as x increases, y stays the same |
|
|
|
244 |
|
00:18:22,220 --> 00:18:26,140 |
|
position, it means there is no linear relationship |
|
|
|
245 |
|
00:18:26,140 --> 00:18:28,900 |
|
between the two variables. So based on the scatter |
|
|
|
246 |
|
00:18:28,900 --> 00:18:33,240 |
|
plot you can have an idea about the relationship |
|
|
|
247 |
|
00:18:33,240 --> 00:18:37,800 |
|
between the two variables. Here I will give a |
|
|
|
248 |
|
00:18:37,800 --> 00:18:41,120 |
|
simple example in order to determine the |
|
|
|
249 |
|
00:18:41,120 --> 00:18:45,160 |
|
correlation coefficient. A real estate agent |
|
|
|
250 |
|
00:18:45,160 --> 00:18:50,380 |
|
wishes to examine the relationship between selling |
|
|
|
251 |
|
00:18:50,380 --> 00:18:54,580 |
|
the price of a home and its size measured in |
|
|
|
252 |
|
00:18:54,580 --> 00:18:57,140 |
|
square feet. So in this case, there are two |
|
|
|
253 |
|
00:18:57,140 --> 00:19:02,400 |
|
variables of interest. One is called selling price |
|
|
|
254 |
|
00:19:02,400 --> 00:19:13,720 |
|
of a home. So here, selling price of a home and |
|
|
|
255 |
|
00:19:13,720 --> 00:19:18,020 |
|
its size. Now, selling price in $1,000. |
|
|
|
256 |
|
00:19:25,360 --> 00:19:29,380 |
|
And size in feet squared. Here we have to |
|
|
|
257 |
|
00:19:29,380 --> 00:19:35,640 |
|
distinguish between dependent and independent. So |
|
|
|
258 |
|
00:19:35,640 --> 00:19:39,740 |
|
your dependent variable is house price, sometimes |
|
|
|
259 |
|
00:19:39,740 --> 00:19:41,620 |
|
called response variable. |
|
|
|
260 |
|
00:19:45,750 --> 00:19:49,490 |
|
The independent variable is the size, which is in |
|
|
|
261 |
|
00:19:49,490 --> 00:19:54,570 |
|
square feet, sometimes called sub-planetary |
|
|
|
262 |
|
00:19:54,570 --> 00:19:54,850 |
|
variable. |
|
|
|
263 |
|
00:19:59,570 --> 00:20:06,370 |
|
So my Y is ceiling rise, and size is square feet, |
|
|
|
264 |
|
00:20:07,530 --> 00:20:12,910 |
|
or size of the house. In this case, there are 10. |
|
|
|
265 |
|
00:20:14,290 --> 00:20:17,890 |
|
It's sample size is 10. So the first house with |
|
|
|
266 |
|
00:20:17,890 --> 00:20:26,850 |
|
size 1,400 square feet, it's selling price is 245 |
|
|
|
267 |
|
00:20:26,850 --> 00:20:31,670 |
|
multiplied by 1,000. Because these values are in |
|
|
|
268 |
|
00:20:31,670 --> 00:20:37,950 |
|
$1,000. Now based on this data, you can first plot |
|
|
|
269 |
|
00:20:37,950 --> 00:20:46,590 |
|
the scatterplot of house price In Y direction, the |
|
|
|
270 |
|
00:20:46,590 --> 00:20:51,870 |
|
vertical direction. So here is house. And rise. |
|
|
|
271 |
|
00:20:54,230 --> 00:21:01,470 |
|
And size in the X axis. You will get this scatter |
|
|
|
272 |
|
00:21:01,470 --> 00:21:07,370 |
|
plot. Now, the data here is just 10 points, so |
|
|
|
273 |
|
00:21:07,370 --> 00:21:12,590 |
|
sometimes it's hard to tell. the relationship |
|
|
|
274 |
|
00:21:12,590 --> 00:21:15,510 |
|
between the two variables if your data is small. |
|
|
|
275 |
|
00:21:16,510 --> 00:21:21,170 |
|
But just this example for illustration. But at |
|
|
|
276 |
|
00:21:21,170 --> 00:21:25,370 |
|
least you can determine that there exists linear |
|
|
|
277 |
|
00:21:25,370 --> 00:21:28,810 |
|
relationship between the two variables. It is |
|
|
|
278 |
|
00:21:28,810 --> 00:21:35,490 |
|
positive. So the form is linear. Direction is |
|
|
|
279 |
|
00:21:35,490 --> 00:21:41,880 |
|
positive. Weak or strong or moderate. Sometimes |
|
|
|
280 |
|
00:21:41,880 --> 00:21:45,620 |
|
it's not easy to tell if it is strong or moderate. |
|
|
|
281 |
|
00:21:47,720 --> 00:21:50,120 |
|
Now if you look at these points, some of them are |
|
|
|
282 |
|
00:21:50,120 --> 00:21:53,700 |
|
close to the straight line and others are away |
|
|
|
283 |
|
00:21:53,700 --> 00:21:56,700 |
|
from the straight line. So maybe there exists |
|
|
|
284 |
|
00:21:56,700 --> 00:22:02,720 |
|
moderate for example, but you cannot say strong. |
|
|
|
285 |
|
00:22:03,930 --> 00:22:08,210 |
|
Here, strong it means the points are close to the |
|
|
|
286 |
|
00:22:08,210 --> 00:22:11,890 |
|
straight line. Sometimes it's hard to tell the |
|
|
|
287 |
|
00:22:11,890 --> 00:22:15,230 |
|
strength of the relationship, but you can know the |
|
|
|
288 |
|
00:22:15,230 --> 00:22:20,990 |
|
form or the direction. But to measure the exact |
|
|
|
289 |
|
00:22:20,990 --> 00:22:24,130 |
|
strength, you have to measure the correlation |
|
|
|
290 |
|
00:22:24,130 --> 00:22:29,810 |
|
coefficient, R. Now, by looking at the data, you |
|
|
|
291 |
|
00:22:29,810 --> 00:22:31,430 |
|
can compute |
|
|
|
292 |
|
00:22:33,850 --> 00:22:42,470 |
|
The sum of x values, y values, sum of x squared, |
|
|
|
293 |
|
00:22:43,290 --> 00:22:48,170 |
|
sum of y squared, also sum of xy. Now plug these |
|
|
|
294 |
|
00:22:48,170 --> 00:22:50,610 |
|
values into the formula we have for the shortcut |
|
|
|
295 |
|
00:22:50,610 --> 00:22:58,210 |
|
formula. You will get R to be 0.76 around 76. |
|
|
|
296 |
|
00:23:04,050 --> 00:23:10,170 |
|
So there exists positive, moderate relationship |
|
|
|
297 |
|
00:23:10,170 --> 00:23:13,770 |
|
between selling |
|
|
|
298 |
|
00:23:13,770 --> 00:23:19,850 |
|
price of a home and its size. So that means if the |
|
|
|
299 |
|
00:23:19,850 --> 00:23:24,670 |
|
size increases, the selling price also increases. |
|
|
|
300 |
|
00:23:25,310 --> 00:23:29,550 |
|
So there exists positive relationship between the |
|
|
|
301 |
|
00:23:29,550 --> 00:23:30,310 |
|
two variables. |
|
|
|
302 |
|
00:23:35,800 --> 00:23:40,300 |
|
Strong it means close to 1, 0.8, 0.85, 0.9, you |
|
|
|
303 |
|
00:23:40,300 --> 00:23:44,400 |
|
can say there exists strong. But fields is not |
|
|
|
304 |
|
00:23:44,400 --> 00:23:47,960 |
|
strong relationship, you can say it's moderate |
|
|
|
305 |
|
00:23:47,960 --> 00:23:53,440 |
|
relationship. Because it's close if now if you |
|
|
|
306 |
|
00:23:53,440 --> 00:23:57,080 |
|
just compare this value and other data gives 9%. |
|
|
|
307 |
|
00:23:58,830 --> 00:24:03,790 |
|
Other one gives 85%. So these values are much |
|
|
|
308 |
|
00:24:03,790 --> 00:24:08,550 |
|
closer to 1 than 0.7, but still this value is |
|
|
|
309 |
|
00:24:08,550 --> 00:24:09,570 |
|
considered to be high. |
|
|
|
310 |
|
00:24:15,710 --> 00:24:16,810 |
|
Any question? |
|
|
|
311 |
|
00:24:19,850 --> 00:24:22,810 |
|
Next, I will give some introduction to regression |
|
|
|
312 |
|
00:24:22,810 --> 00:24:23,390 |
|
analysis. |
|
|
|
313 |
|
00:24:26,970 --> 00:24:32,210 |
|
Regression analysis used to number one, predict |
|
|
|
314 |
|
00:24:32,210 --> 00:24:35,050 |
|
the value of a dependent variable based on the |
|
|
|
315 |
|
00:24:35,050 --> 00:24:39,250 |
|
value of at least one independent variable. So by |
|
|
|
316 |
|
00:24:39,250 --> 00:24:42,490 |
|
using the data we have for selling price of a home |
|
|
|
317 |
|
00:24:42,490 --> 00:24:48,370 |
|
and size, you can predict the selling price by |
|
|
|
318 |
|
00:24:48,370 --> 00:24:51,510 |
|
knowing the value of its size. So suppose for |
|
|
|
319 |
|
00:24:51,510 --> 00:24:54,870 |
|
example, You know that the size of a house is |
|
|
|
320 |
|
00:24:54,870 --> 00:25:03,510 |
|
1450, 1450 square feet. What do you predict its |
|
|
|
321 |
|
00:25:03,510 --> 00:25:10,190 |
|
size, its sale or price? So by using this value, |
|
|
|
322 |
|
00:25:10,310 --> 00:25:16,510 |
|
we can predict the selling price. Next, explain |
|
|
|
323 |
|
00:25:16,510 --> 00:25:19,890 |
|
the impact of changes in independent variable on |
|
|
|
324 |
|
00:25:19,890 --> 00:25:23,270 |
|
the dependent variable. You can say, for example, |
|
|
|
325 |
|
00:25:23,510 --> 00:25:30,650 |
|
90% of the variability in the dependent variable |
|
|
|
326 |
|
00:25:30,650 --> 00:25:36,790 |
|
in selling price is explained by its size. So we |
|
|
|
327 |
|
00:25:36,790 --> 00:25:39,410 |
|
can predict the value of dependent variable based |
|
|
|
328 |
|
00:25:39,410 --> 00:25:42,890 |
|
on a value of one independent variable at least. |
|
|
|
329 |
|
00:25:43,870 --> 00:25:47,090 |
|
Or also explain the impact of changes in |
|
|
|
330 |
|
00:25:47,090 --> 00:25:49,550 |
|
independent variable on the dependent variable. |
|
|
|
331 |
|
00:25:51,420 --> 00:25:53,920 |
|
Sometimes there exists more than one independent |
|
|
|
332 |
|
00:25:53,920 --> 00:25:59,680 |
|
variable. For example, maybe there are more than |
|
|
|
333 |
|
00:25:59,680 --> 00:26:04,500 |
|
one variable that affects a price, a selling |
|
|
|
334 |
|
00:26:04,500 --> 00:26:10,300 |
|
price. For example, beside selling |
|
|
|
335 |
|
00:26:10,300 --> 00:26:16,280 |
|
price, beside size, maybe location. |
|
|
|
336 |
|
00:26:19,480 --> 00:26:23,580 |
|
Maybe location is also another factor that affects |
|
|
|
337 |
|
00:26:23,580 --> 00:26:27,360 |
|
the selling price. So in this case there are two |
|
|
|
338 |
|
00:26:27,360 --> 00:26:32,240 |
|
variables. If there exists more than one variable, |
|
|
|
339 |
|
00:26:32,640 --> 00:26:36,080 |
|
in this case we have something called multiple |
|
|
|
340 |
|
00:26:36,080 --> 00:26:38,680 |
|
linear regression. |
|
|
|
341 |
|
00:26:42,030 --> 00:26:46,710 |
|
Here, we just talk about one independent variable. |
|
|
|
342 |
|
00:26:47,030 --> 00:26:51,610 |
|
There is only, in this chapter, there is only one |
|
|
|
343 |
|
00:26:51,610 --> 00:26:58,330 |
|
x. So it's called simple linear |
|
|
|
344 |
|
00:26:58,330 --> 00:26:59,330 |
|
regression. |
|
|
|
345 |
|
00:27:02,190 --> 00:27:07,930 |
|
The calculations for multiple takes time. So we |
|
|
|
346 |
|
00:27:07,930 --> 00:27:11,430 |
|
are going just to cover one independent variable. |
|
|
|
347 |
|
00:27:11,930 --> 00:27:14,290 |
|
But if there exists more than one, in this case |
|
|
|
348 |
|
00:27:14,290 --> 00:27:18,250 |
|
you have to use some statistical software as SPSS. |
|
|
|
349 |
|
00:27:18,470 --> 00:27:23,390 |
|
Because in that case you can just select a |
|
|
|
350 |
|
00:27:23,390 --> 00:27:25,970 |
|
regression analysis from SPSS, then you can run |
|
|
|
351 |
|
00:27:25,970 --> 00:27:28,590 |
|
the multiple regression without doing any |
|
|
|
352 |
|
00:27:28,590 --> 00:27:34,190 |
|
computations. But here we just covered one |
|
|
|
353 |
|
00:27:34,190 --> 00:27:36,820 |
|
independent variable. In this case, it's called |
|
|
|
354 |
|
00:27:36,820 --> 00:27:41,980 |
|
simple linear regression. Again, the dependent |
|
|
|
355 |
|
00:27:41,980 --> 00:27:44,600 |
|
variable is the variable we wish to predict or |
|
|
|
356 |
|
00:27:44,600 --> 00:27:50,020 |
|
explain, the same as weight. Independent variable, |
|
|
|
357 |
|
00:27:50,180 --> 00:27:52,440 |
|
the variable used to predict or explain the |
|
|
|
358 |
|
00:27:52,440 --> 00:27:54,000 |
|
dependent variable. |
|
|
|
359 |
|
00:27:57,400 --> 00:28:00,540 |
|
For simple linear regression model, there is only |
|
|
|
360 |
|
00:28:00,540 --> 00:28:01,800 |
|
one independent variable. |
|
|
|
361 |
|
00:28:04,830 --> 00:28:08,450 |
|
Another example for simple linear regression. |
|
|
|
362 |
|
00:28:08,770 --> 00:28:11,590 |
|
Suppose we are talking about your scores. |
|
|
|
363 |
|
00:28:14,210 --> 00:28:17,770 |
|
Scores is the dependent variable can be affected |
|
|
|
364 |
|
00:28:17,770 --> 00:28:21,050 |
|
by number of hours. |
|
|
|
365 |
|
00:28:25,130 --> 00:28:31,030 |
|
Hour of study. Number of studying hours. |
|
|
|
366 |
|
00:28:36,910 --> 00:28:39,810 |
|
Maybe as number of studying hour increases, your |
|
|
|
367 |
|
00:28:39,810 --> 00:28:43,390 |
|
scores also increase. In this case, if there is |
|
|
|
368 |
|
00:28:43,390 --> 00:28:46,330 |
|
only one X, one independent variable, it's called |
|
|
|
369 |
|
00:28:46,330 --> 00:28:51,110 |
|
simple linear regression. Maybe another variable, |
|
|
|
370 |
|
00:28:52,270 --> 00:28:59,730 |
|
number of missing classes or |
|
|
|
371 |
|
00:28:59,730 --> 00:29:03,160 |
|
attendance. As number of missing classes |
|
|
|
372 |
|
00:29:03,160 --> 00:29:06,380 |
|
increases, your score goes down. That means there |
|
|
|
373 |
|
00:29:06,380 --> 00:29:09,400 |
|
exists negative relationship between missing |
|
|
|
374 |
|
00:29:09,400 --> 00:29:13,540 |
|
classes and your score. So sometimes, maybe there |
|
|
|
375 |
|
00:29:13,540 --> 00:29:16,580 |
|
exists positive or negative. It depends on the |
|
|
|
376 |
|
00:29:16,580 --> 00:29:20,040 |
|
variable itself. In this case, if there are more |
|
|
|
377 |
|
00:29:20,040 --> 00:29:23,180 |
|
than one variable, then we are talking about |
|
|
|
378 |
|
00:29:23,180 --> 00:29:28,300 |
|
multiple linear regression model. But here, we |
|
|
|
379 |
|
00:29:28,300 --> 00:29:33,630 |
|
have only one independent variable. In addition to |
|
|
|
380 |
|
00:29:33,630 --> 00:29:37,230 |
|
that, a relationship between x and y is described |
|
|
|
381 |
|
00:29:37,230 --> 00:29:40,850 |
|
by a linear function. So there exists a straight |
|
|
|
382 |
|
00:29:40,850 --> 00:29:46,270 |
|
line between the two variables. The changes in y |
|
|
|
383 |
|
00:29:46,270 --> 00:29:50,210 |
|
are assumed to be related to changes in x only. So |
|
|
|
384 |
|
00:29:50,210 --> 00:29:54,270 |
|
any change in y is related only to changes in x. |
|
|
|
385 |
|
00:29:54,730 --> 00:29:57,810 |
|
So that's the simple case we have for regression, |
|
|
|
386 |
|
00:29:58,890 --> 00:30:01,170 |
|
that we have only one independent |
|
|
|
387 |
|
00:30:03,890 --> 00:30:07,070 |
|
Variable. Types of relationships, as we mentioned, |
|
|
|
388 |
|
00:30:07,210 --> 00:30:12,190 |
|
maybe there exist linear, it means there exist |
|
|
|
389 |
|
00:30:12,190 --> 00:30:16,490 |
|
straight line between X and Y, either linear |
|
|
|
390 |
|
00:30:16,490 --> 00:30:22,050 |
|
positive or negative, or sometimes there exist non |
|
|
|
391 |
|
00:30:22,050 --> 00:30:25,830 |
|
-linear relationship, it's called curved linear |
|
|
|
392 |
|
00:30:25,830 --> 00:30:29,290 |
|
relationship. The same as this one, it's parabola. |
|
|
|
393 |
|
00:30:32,570 --> 00:30:35,150 |
|
Now in this case there is no linear relationship |
|
|
|
394 |
|
00:30:35,150 --> 00:30:39,690 |
|
but there exists curved linear or something like |
|
|
|
395 |
|
00:30:39,690 --> 00:30:45,910 |
|
this one. So these types of non-linear |
|
|
|
396 |
|
00:30:45,910 --> 00:30:49,530 |
|
relationship between the two variables. Here we |
|
|
|
397 |
|
00:30:49,530 --> 00:30:54,070 |
|
are covering just the linear relationship between |
|
|
|
398 |
|
00:30:54,070 --> 00:30:56,570 |
|
the two variables. So based on the scatter plot |
|
|
|
399 |
|
00:30:56,570 --> 00:31:00,620 |
|
you can determine the direction. The form, the |
|
|
|
400 |
|
00:31:00,620 --> 00:31:03,860 |
|
strength. Here, the form we are talking about is |
|
|
|
401 |
|
00:31:03,860 --> 00:31:04,720 |
|
just linear. |
|
|
|
402 |
|
00:31:08,700 --> 00:31:13,260 |
|
Now, another type of relationship, the strength of |
|
|
|
403 |
|
00:31:13,260 --> 00:31:16,940 |
|
the relationship. Here, the points, either for |
|
|
|
404 |
|
00:31:16,940 --> 00:31:20,570 |
|
this graph or the other one, These points are |
|
|
|
405 |
|
00:31:20,570 --> 00:31:24,570 |
|
close to the straight line, it means there exists |
|
|
|
406 |
|
00:31:24,570 --> 00:31:28,210 |
|
strong positive relationship or strong negative |
|
|
|
407 |
|
00:31:28,210 --> 00:31:31,230 |
|
relationship. So it depends on the direction. So |
|
|
|
408 |
|
00:31:31,230 --> 00:31:35,710 |
|
strong either positive or strong negative. Here |
|
|
|
409 |
|
00:31:35,710 --> 00:31:38,850 |
|
the points are scattered away from the regression |
|
|
|
410 |
|
00:31:38,850 --> 00:31:41,790 |
|
line, so you can say there exists weak |
|
|
|
411 |
|
00:31:41,790 --> 00:31:45,090 |
|
relationship, either weak positive or weak |
|
|
|
412 |
|
00:31:45,090 --> 00:31:49,650 |
|
negative. It depends on the direction of the |
|
|
|
413 |
|
00:31:49,650 --> 00:31:54,270 |
|
relationship between the two variables. Sometimes |
|
|
|
414 |
|
00:31:54,270 --> 00:31:59,680 |
|
there is no relationship or actually there is no |
|
|
|
415 |
|
00:31:59,680 --> 00:32:02,340 |
|
linear relationship between the two variables. If |
|
|
|
416 |
|
00:32:02,340 --> 00:32:05,660 |
|
the points are scattered away from the regression |
|
|
|
417 |
|
00:32:05,660 --> 00:32:09,800 |
|
line, I mean you cannot determine if it is |
|
|
|
418 |
|
00:32:09,800 --> 00:32:13,160 |
|
positive or negative, then there is no |
|
|
|
419 |
|
00:32:13,160 --> 00:32:16,220 |
|
relationship between the two variables, the same |
|
|
|
420 |
|
00 |
|
|
|
445 |
|
00:34:40,270 --> 00:34:43,850 |
|
could be negative or |
|
|
|
446 |
|
00:34:46,490 --> 00:34:49,350 |
|
Maybe the straight line passes through the origin |
|
|
|
447 |
|
00:34:49,350 --> 00:34:56,990 |
|
point. So in this case, beta zero equals zero. So |
|
|
|
448 |
|
00:34:56,990 --> 00:34:59,890 |
|
it could be positive and negative or equal zero, |
|
|
|
449 |
|
00:35:00,430 --> 00:35:05,510 |
|
but still we have positive relationship. That |
|
|
|
450 |
|
00:35:05,510 --> 00:35:09,970 |
|
means the value of beta zero, the sign of beta |
|
|
|
451 |
|
00:35:09,970 --> 00:35:13,310 |
|
zero does not affect the relationship between Y |
|
|
|
452 |
|
00:35:13,310 --> 00:35:17,850 |
|
and X. Because here in the three cases, there |
|
|
|
453 |
|
00:35:17,850 --> 00:35:22,390 |
|
exists positive relationship, but beta zero could |
|
|
|
454 |
|
00:35:22,390 --> 00:35:25,370 |
|
be positive or negative or equal zero, but still |
|
|
|
455 |
|
00:35:25,370 --> 00:35:31,720 |
|
we have positive relationship. I mean, you cannot |
|
|
|
456 |
|
00:35:31,720 --> 00:35:35,060 |
|
determine by looking at beta 0, you cannot |
|
|
|
457 |
|
00:35:35,060 --> 00:35:37,940 |
|
determine if there is a positive or negative |
|
|
|
458 |
|
00:35:37,940 --> 00:35:41,720 |
|
relationship. The other term is beta 1. Beta 1 is |
|
|
|
459 |
|
00:35:41,720 --> 00:35:46,900 |
|
the population slope coefficient. Now, the sign of |
|
|
|
460 |
|
00:35:46,900 --> 00:35:50,010 |
|
the slope determines the direction of the |
|
|
|
461 |
|
00:35:50,010 --> 00:35:54,090 |
|
relationship. That means if the slope has positive |
|
|
|
462 |
|
00:35:54,090 --> 00:35:56,570 |
|
sign, it means there exists positive relationship. |
|
|
|
463 |
|
00:35:57,330 --> 00:35:59,370 |
|
Otherwise if it is negative, then there is |
|
|
|
464 |
|
00:35:59,370 --> 00:36:01,390 |
|
negative relationship between the two variables. |
|
|
|
465 |
|
00:36:02,130 --> 00:36:05,310 |
|
So the sign of the slope determines the direction. |
|
|
|
466 |
|
00:36:06,090 --> 00:36:11,290 |
|
But the sign of beta zero has no meaning about the |
|
|
|
467 |
|
00:36:11,290 --> 00:36:15,470 |
|
relationship between Y and X. X is your |
|
|
|
468 |
|
00:36:15,470 --> 00:36:19,630 |
|
independent variable, Y is your independent |
|
|
|
469 |
|
00:36:19,630 --> 00:36:19,650 |
|
your independent variable, Y is your independent |
|
|
|
470 |
|
00:36:19,650 --> 00:36:21,250 |
|
variable, Y is your independent variable, Y is |
|
|
|
471 |
|
00:36:21,250 --> 00:36:24,370 |
|
variable, Y is your independent variable, Y is |
|
|
|
472 |
|
00:36:24,370 --> 00:36:24,430 |
|
variable, Y is your independent variable, Y is |
|
|
|
473 |
|
00:36:24,430 --> 00:36:24,770 |
|
your independent variable, Y is your independent |
|
|
|
474 |
|
00:36:24,770 --> 00:36:27,490 |
|
variable, Y is your independent variable, Y is |
|
|
|
475 |
|
00:36:27,490 --> 00:36:30,110 |
|
your independent variable, Y is your It means |
|
|
|
476 |
|
00:36:30,110 --> 00:36:32,450 |
|
there are some errors you don't know about it |
|
|
|
477 |
|
00:36:32,450 --> 00:36:36,130 |
|
because you ignore some other variables that may |
|
|
|
478 |
|
00:36:36,130 --> 00:36:39,410 |
|
affect the selling price. Maybe you select a |
|
|
|
479 |
|
00:36:39,410 --> 00:36:42,490 |
|
random sample, that sample is small. Maybe there |
|
|
|
480 |
|
00:36:42,490 --> 00:36:46,270 |
|
is a random, I'm sorry, there is sampling error. |
|
|
|
481 |
|
00:36:47,070 --> 00:36:52,980 |
|
So all of these are called random error term. So |
|
|
|
482 |
|
00:36:52,980 --> 00:36:57,420 |
|
all of them are in this term. So epsilon I means |
|
|
|
483 |
|
00:36:57,420 --> 00:37:00,340 |
|
something you don't include in your regression |
|
|
|
484 |
|
00:37:00,340 --> 00:37:03,280 |
|
modeling. For example, you don't include all the |
|
|
|
485 |
|
00:37:03,280 --> 00:37:06,180 |
|
independent variables that affect Y, or your |
|
|
|
486 |
|
00:37:06,180 --> 00:37:09,700 |
|
sample size is not large enough. So all of these |
|
|
|
487 |
|
00:37:09,700 --> 00:37:14,260 |
|
measured in random error term. So epsilon I is |
|
|
|
488 |
|
00:37:14,260 --> 00:37:18,840 |
|
random error component, beta 0 plus beta 1X is |
|
|
|
489 |
|
00:37:18,840 --> 00:37:25,070 |
|
called linear component. So that's the simple |
|
|
|
490 |
|
00:37:25,070 --> 00:37:31,430 |
|
linear regression model. Now, the data you have, |
|
|
|
491 |
|
00:37:32,850 --> 00:37:38,210 |
|
the blue circles represent the observed value. So |
|
|
|
492 |
|
00:37:38,210 --> 00:37:47,410 |
|
these blue circles are the observed values. So we |
|
|
|
493 |
|
00:37:47,410 --> 00:37:49,370 |
|
have observed. |
|
|
|
494 |
|
00:37:52,980 --> 00:37:57,940 |
|
Y observed value of Y for each value X. The |
|
|
|
495 |
|
00:37:57,940 --> 00:38:03,360 |
|
regression line is the blue, the red one. It's |
|
|
|
496 |
|
00:38:03,360 --> 00:38:07,560 |
|
called the predicted values. Predicted Y. |
|
|
|
497 |
|
00:38:08,180 --> 00:38:14,760 |
|
Predicted Y is denoted always by Y hat. Now the |
|
|
|
498 |
|
00:38:14,760 --> 00:38:19,740 |
|
difference between Y and Y hat. It's called the |
|
|
|
499 |
|
00:38:19,740 --> 00:38:20,200 |
|
error term. |
|
|
|
500 |
|
00:38:24,680 --> 00:38:28,000 |
|
It's actually the difference between the observed |
|
|
|
501 |
|
00:38:28,000 --> 00:38:31,600 |
|
value and its predicted value. Now, the predicted |
|
|
|
502 |
|
00:38:31,600 --> 00:38:34,720 |
|
value can be determined by using the regression |
|
|
|
503 |
|
00:38:34,720 --> 00:38:39,180 |
|
line. So this line is the predicted value of Y for |
|
|
|
504 |
|
00:38:39,180 --> 00:38:44,480 |
|
XR. Again, beta zero is the intercept. As we |
|
|
|
505 |
|
00:38:44,480 --> 00:38:46,260 |
|
mentioned before, it could be positive or negative |
|
|
|
506 |
|
00:38:46,260 --> 00:38:52,600 |
|
or even equal zero. The slope is changing Y. |
|
|
|
507 |
|
00:38:55,140 --> 00:38:57,580 |
|
Divide by change of x. |
|
|
|
508 |
|
00:39:01,840 --> 00:39:07,140 |
|
So these are the components for the simple linear |
|
|
|
509 |
|
00:39:07,140 --> 00:39:10,840 |
|
regression model. Y again represents the |
|
|
|
510 |
|
00:39:10,840 --> 00:39:14,960 |
|
independent variable. Beta 0 y intercept. Beta 1 |
|
|
|
511 |
|
00:39:14,960 --> 00:39:17,960 |
|
is your slope. And the slope determines the |
|
|
|
512 |
|
00:39:17,960 --> 00:39:20,900 |
|
direction of the relationship. X independent |
|
|
|
513 |
|
00:39:20,900 --> 00:39:25,270 |
|
variable epsilon i is the random error term. Any |
|
|
|
514 |
|
00:39:25,270 --> 00:39:25,650 |
|
question? |
|
|
|
515 |
|
00:39:31,750 --> 00:39:36,610 |
|
The relationship may be positive or negative. It |
|
|
|
516 |
|
00:39:36,610 --> 00:39:37,190 |
|
could be negative. |
|
|
|
517 |
|
00:39:40,950 --> 00:39:42,710 |
|
Now, for negative relationship, |
|
|
|
518 |
|
00:39:57,000 --> 00:40:04,460 |
|
Or negative, where beta zero is negative. |
|
|
|
519 |
|
00:40:04,520 --> 00:40:08,700 |
|
Or beta |
|
|
|
520 |
|
00:40:08,700 --> 00:40:09,740 |
|
zero equals zero. |
|
|
|
521 |
|
00:40:16,680 --> 00:40:20,620 |
|
So here there exists negative relationship, but |
|
|
|
522 |
|
00:40:20,620 --> 00:40:22,060 |
|
beta zero may be positive. |
|
|
|
523 |
|
00:40:25,870 --> 00:40:30,210 |
|
So again, the sign of beta 0 also does not affect |
|
|
|
524 |
|
00:40:30,210 --> 00:40:31,990 |
|
the relationship between the two variables. |
|
|
|
525 |
|
00:40:36,230 --> 00:40:40,590 |
|
Now, we don't actually know the values of beta 0 |
|
|
|
526 |
|
00:40:40,590 --> 00:40:44,510 |
|
and beta 1. We are going to estimate these values |
|
|
|
527 |
|
00:40:44,510 --> 00:40:48,110 |
|
from the sample we have. So the simple linear |
|
|
|
528 |
|
00:40:48,110 --> 00:40:50,970 |
|
regression equation provides an estimate of the |
|
|
|
529 |
|
00:40:50,970 --> 00:40:55,270 |
|
population regression line. So here we have Yi hat |
|
|
|
530 |
|
00:40:55,270 --> 00:41:00,010 |
|
is the estimated or predicted Y value for |
|
|
|
531 |
|
00:41:00,010 --> 00:41:00,850 |
|
observation I. |
|
|
|
532 |
|
00:41:03,530 --> 00:41:08,220 |
|
The estimate of the regression intercept P0. The |
|
|
|
533 |
|
00:41:08,220 --> 00:41:11,360 |
|
estimate of the regression slope is b1, and this |
|
|
|
534 |
|
00:41:11,360 --> 00:41:16,680 |
|
is your x, all independent variable. So here is |
|
|
|
535 |
|
00:41:16,680 --> 00:41:20,340 |
|
the regression equation. Simple linear regression |
|
|
|
536 |
|
00:41:20,340 --> 00:41:24,400 |
|
equation is given by y hat, the predicted value of |
|
|
|
537 |
|
00:41:24,400 --> 00:41:29,380 |
|
y equals b0 plus b1 times x1. |
|
|
|
538 |
|
00:41:31,240 --> 00:41:35,960 |
|
Now these coefficients, b0 and b1 can be computed |
|
|
|
539 |
|
00:41:37,900 --> 00:41:43,040 |
|
by the following equations. So the regression |
|
|
|
540 |
|
00:41:43,040 --> 00:41:52,920 |
|
equation is |
|
|
|
541 |
|
00:41:52,920 --> 00:41:57,260 |
|
given by y hat equals b0 plus b1x. |
|
|
|
542 |
|
00:41:59,940 --> 00:42:06,140 |
|
Now the slope, b1, is r times standard deviation |
|
|
|
543 |
|
00:42:06,140 --> 00:42:10,540 |
|
of y Times standard deviation of x. This is the |
|
|
|
544 |
|
00:42:10,540 --> 00:42:13,820 |
|
simplest equation to determine the value of the |
|
|
|
545 |
|
00:42:13,820 --> 00:42:18,980 |
|
star. B1r, r is the correlation coefficient. Sy is |
|
|
|
546 |
|
00:42:18,980 --> 00:42:25,080 |
|
xr, the standard deviations of y and x. Where b0, |
|
|
|
547 |
|
00:42:25,520 --> 00:42:30,880 |
|
which is y intercept, is y bar minus b x bar, or |
|
|
|
548 |
|
00:42:30,880 --> 00:42:38,100 |
|
b1 x bar. Sx, as we know, is the sum of x minus y |
|
|
|
549 |
|
00:42:38,100 --> 00:42:40,460 |
|
squared divided by n minus 1 under square root, |
|
|
|
550 |
|
00:42:40,900 --> 00:42:47,060 |
|
similarly for y values. So this, how can we, these |
|
|
|
551 |
|
00:42:47,060 --> 00:42:52,380 |
|
formulas compute the values of b0 and b1. So we |
|
|
|
552 |
|
00:42:52,380 --> 00:42:54,600 |
|
are going to use these equations in order to |
|
|
|
553 |
|
00:42:54,600 --> 00:42:58,960 |
|
determine the values of b0 and b1. |
|
|
|
554 |
|
00:43:04,670 --> 00:43:07,710 |
|
Now, what's your interpretation about the slope |
|
|
|
555 |
|
00:43:07,710 --> 00:43:13,130 |
|
and the intercept? For example, suppose we are |
|
|
|
556 |
|
00:43:13,130 --> 00:43:18,610 |
|
talking about your score Y and |
|
|
|
557 |
|
00:43:18,610 --> 00:43:22,110 |
|
X number of missing classes. |
|
|
|
558 |
|
00:43:29,210 --> 00:43:35,460 |
|
And suppose, for example, Y hat Equal 95 minus 5x. |
|
|
|
559 |
|
00:43:37,780 --> 00:43:41,420 |
|
Now let's see what's the interpretation of B0. |
|
|
|
560 |
|
00:43:42,300 --> 00:43:45,060 |
|
This is B0. So B0 is 95. |
|
|
|
561 |
|
00:43:47,660 --> 00:43:51,960 |
|
And B1 is 5. Now what's your interpretation about |
|
|
|
562 |
|
00:43:51,960 --> 00:43:57,740 |
|
B0 and B1? B0 is the estimated mean value of Y |
|
|
|
563 |
|
00:43:57,740 --> 00:44:02,560 |
|
when the value of X is 0. that means if the |
|
|
|
564 |
|
00:44:02,560 --> 00:44:08,500 |
|
student does not miss any class that means x |
|
|
|
565 |
|
00:44:08,500 --> 00:44:13,260 |
|
equals zero in this case we predict or we estimate |
|
|
|
566 |
|
00:44:13,260 --> 00:44:19,880 |
|
the mean value of his score or her score is 95 so |
|
|
|
567 |
|
00:44:19,880 --> 00:44:27,500 |
|
95 it means when x is zero if x is zero then we |
|
|
|
568 |
|
00:44:27,500 --> 00:44:35,350 |
|
expect his or Here, the score is 95. So that means |
|
|
|
569 |
|
00:44:35,350 --> 00:44:39,830 |
|
B0 is the estimated mean value of Y when the value |
|
|
|
570 |
|
00:44:39,830 --> 00:44:40,630 |
|
of X is 0. |
|
|
|
571 |
|
00:44:43,370 --> 00:44:46,590 |
|
Now, what's the meaning of the slope? The slope in |
|
|
|
572 |
|
00:44:46,590 --> 00:44:51,290 |
|
this case is negative Y. B1, which is the slope, |
|
|
|
573 |
|
00:44:51,590 --> 00:44:57,610 |
|
is the estimated change in the mean of Y. as a |
|
|
|
574 |
|
00:44:57,610 --> 00:45:03,050 |
|
result of a one unit change in x for example let's |
|
|
|
575 |
|
00:45:03,050 --> 00:45:07,070 |
|
compute y for different values of x suppose x is |
|
|
|
576 |
|
00:45:07,070 --> 00:45:15,510 |
|
one now we predict his score to be 95 minus 5 |
|
|
|
577 |
|
00:45:15,510 --> 00:45:25,470 |
|
times 1 which is 90 when x is 2 for example Y hat |
|
|
|
578 |
|
00:45:25,470 --> 00:45:28,570 |
|
is 95 minus 5 times 2, so that's 85. |
|
|
|
579 |
|
00:45:31,950 --> 00:45:39,970 |
|
So for each one unit, there is a drop by five |
|
|
|
580 |
|
00:45:39,970 --> 00:45:43,750 |
|
units in his score. That means if number of |
|
|
|
581 |
|
00:45:43,750 --> 00:45:47,550 |
|
missing classes increases by one unit, then his or |
|
|
|
582 |
|
00:45:47,550 --> 00:45:51,790 |
|
her weight is expected to be reduced by five units |
|
|
|
583 |
|
00:45:51,790 --> 00:45:56,150 |
|
because the sign is negative. another example |
|
|
|
584 |
|
00:45:56,150 --> 00:46:05,910 |
|
suppose again we are interested in whales and |
|
|
|
585 |
|
00:46:05,910 --> 00:46:16,170 |
|
angels and imagine that just |
|
|
|
586 |
|
00:46:16,170 --> 00:46:21,670 |
|
for example y equal y hat equals three plus four x |
|
|
|
587 |
|
00:46:21,670 --> 00:46:29,830 |
|
now y hat equals 3 if x equals zero. That has no |
|
|
|
588 |
|
00:46:29,830 --> 00:46:34,510 |
|
meaning because you cannot say age of zero. So |
|
|
|
589 |
|
00:46:34,510 --> 00:46:40,450 |
|
sometimes the meaning of y intercept does not make |
|
|
|
590 |
|
00:46:40,450 --> 00:46:46,150 |
|
sense because you cannot say x equals zero. Now |
|
|
|
591 |
|
00:46:46,150 --> 00:46:50,690 |
|
for the stock of four, that means as his or her |
|
|
|
592 |
|
00:46:50,690 --> 00:46:55,550 |
|
weight increases by one year, Then we expect his |
|
|
|
593 |
|
00:46:55,550 --> 00:47:00,470 |
|
weight to increase by four kilograms. So as one |
|
|
|
594 |
|
00:47:00,470 --> 00:47:05,130 |
|
unit increase in x, y is our, his weight is |
|
|
|
595 |
|
00:47:05,130 --> 00:47:10,150 |
|
expected to increase by four units. So again, |
|
|
|
596 |
|
00:47:10,370 --> 00:47:16,950 |
|
sometimes we can interpret the y intercept, but in |
|
|
|
597 |
|
00:47:16,950 --> 00:47:18,670 |
|
some cases it has no meaning. |
|
|
|
598 |
|
00:47:24,970 --> 00:47:27,190 |
|
Now for the previous example, for the selling |
|
|
|
599 |
|
00:47:27,190 --> 00:47:32,930 |
|
price of a home and its size, B1rSy divided by Sx, |
|
|
|
600 |
|
00:47:33,790 --> 00:47:43,550 |
|
r is computed, r is found to be 76%, 76%Sy divided |
|
|
|
601 |
|
00:47:43,550 --> 00:47:49,990 |
|
by Sx, that will give 0.109. B0y bar minus B1x |
|
|
|
602 |
|
00:47:49,990 --> 00:47:50,670 |
|
bar, |
|
|
|
603 |
|
00:47:53,610 --> 00:48:00,150 |
|
Y bar for this data is 286 minus D1. So we have to |
|
|
|
604 |
|
00:48:00,150 --> 00:48:03,490 |
|
compute first D1 because we use it in order to |
|
|
|
605 |
|
00:48:03,490 --> 00:48:08,590 |
|
determine D0. And calculation gives 98. So that |
|
|
|
606 |
|
00:48:08,590 --> 00:48:16,450 |
|
means based on these equations, Y hat equals 0 |
|
|
|
607 |
|
00:48:16,450 --> 00:48:22,990 |
|
.10977 plus 98.248. |
|
|
|
608 |
|
00:48:24,790 --> 00:48:29,370 |
|
times X. X is the size. |
|
|
|
609 |
|
00:48:32,890 --> 00:48:39,830 |
|
0.1 B1 |
|
|
|
610 |
|
00:48:39,830 --> 00:48:45,310 |
|
is |
|
|
|
611 |
|
00:48:45,310 --> 00:48:56,650 |
|
0.1, B0 is 98, so 98.248 plus B1. So this is your |
|
|
|
612 |
|
00:48:56,650 --> 00:49:03,730 |
|
regression equation. So again, the intercept is |
|
|
|
613 |
|
00:49:03,730 --> 00:49:09,750 |
|
98. So this amount, the segment is 98. Now the |
|
|
|
614 |
|
00:49:09,750 --> 00:49:14,790 |
|
slope is 0.109. So house price, the expected value |
|
|
|
615 |
|
00:49:14,790 --> 00:49:21,270 |
|
of house price equals B098 plus 0.109 square feet. |
|
|
|
616 |
|
00:49:23,150 --> 00:49:27,630 |
|
So that's the prediction line for the house price. |
|
|
|
617 |
|
00:49:28,510 --> 00:49:34,370 |
|
So again, house price equal B0 98 plus 0.10977 |
|
|
|
618 |
|
00:49:34,370 --> 00:49:36,930 |
|
times square root. Now, what's your interpretation |
|
|
|
619 |
|
00:49:36,930 --> 00:49:41,950 |
|
about B0 and B1? B0 is the estimated mean value of |
|
|
|
620 |
|
00:49:41,950 --> 00:49:46,430 |
|
Y when the value of X is 0. So if X is 0, this |
|
|
|
621 |
|
00:49:46,430 --> 00:49:52,980 |
|
range of X observed X values and you have a home |
|
|
|
622 |
|
00:49:52,980 --> 00:49:57,860 |
|
or a house of size zero. So that means this value |
|
|
|
623 |
|
00:49:57,860 --> 00:50:02,680 |
|
has no meaning. Because a house cannot have a |
|
|
|
624 |
|
00:50:02,680 --> 00:50:06,400 |
|
square footage of zero. So B0 has no practical |
|
|
|
625 |
|
00:50:06,400 --> 00:50:10,040 |
|
application in this case. So sometimes it makes |
|
|
|
626 |
|
00:50:10,040 --> 00:50:17,620 |
|
sense, in other cases it doesn't have that. So for |
|
|
|
627 |
|
00:50:17,620 --> 00:50:21,790 |
|
this specific example, B0 has no practical |
|
|
|
628 |
|
00:50:21,790 --> 00:50:28,210 |
|
application in this case. But B1 which is 0.1097, |
|
|
|
629 |
|
00:50:28,930 --> 00:50:33,050 |
|
B1 estimates the change in the mean value of Y as |
|
|
|
630 |
|
00:50:33,050 --> 00:50:36,730 |
|
a result of one unit increasing X. So for this |
|
|
|
631 |
|
00:50:36,730 --> 00:50:41,640 |
|
value which is 0.109, it means This fellow tells |
|
|
|
632 |
|
00:50:41,640 --> 00:50:46,420 |
|
us that the mean value of a house can increase by |
|
|
|
633 |
|
00:50:46,420 --> 00:50:52,280 |
|
this amount, increase by 0.1097, but we have to |
|
|
|
634 |
|
00:50:52,280 --> 0 |
|
|
|
667 |
|
00:53:54,190 --> 00:53:57,770 |
|
that's |
|
|
|
668 |
|
00:53:57,770 --> 00:53:57,990 |
|
all |
|
|