1
00:00:01,100 --> 00:00:07,850
And this video will assess the accuracy of these coefficients because it can be done that we have calculated.

2
00:00:09,140 --> 00:00:11,750
I will start by restating the situation that we are in.

3
00:00:13,320 --> 00:00:16,380
In the world, millions of House transactions were happening.

4
00:00:16,740 --> 00:00:24,120
We picked a small set or a small sample of 506 such observations and decided to identify the relationship

5
00:00:24,120 --> 00:00:26,160
between house place and number of rooms.

6
00:00:27,510 --> 00:00:33,170
When I use the formula shown in previous video, we get the values of B does yellow and Beethoven.

7
00:00:34,530 --> 00:00:38,820
This line is minimizing the squared error on the sample of five under six points.

8
00:00:40,170 --> 00:00:46,130
If we had selected some other 506 observations, we may have got some other values of B does it when

9
00:00:46,130 --> 00:00:46,650
we do one.

10
00:00:47,730 --> 00:00:54,270
If we take all the million observations of the world and then get a line, this line will be called

11
00:00:54,270 --> 00:00:55,740
the population regression line.

12
00:00:56,610 --> 00:01:00,150
This line may or may not be seen as this simple regression line.

13
00:01:02,070 --> 00:01:07,800
You can see in this graph, this blue line, it is simple regression line, which is minimizing the

14
00:01:07,920 --> 00:01:09,990
error on these data points.

15
00:01:10,140 --> 00:01:11,340
D sample datapoint.

16
00:01:13,370 --> 00:01:20,260
But this red line reality population, the green line, which would have come if I would have thought

17
00:01:20,260 --> 00:01:25,310
the least good at a for all the data points of the population.

18
00:01:26,600 --> 00:01:27,860
These two de Nazeem.

19
00:01:29,610 --> 00:01:33,480
Now, the problem is this, we do not have all the observations.

20
00:01:33,670 --> 00:01:35,830
Therefore, we cannot get population immigration lane.

21
00:01:36,910 --> 00:01:40,420
We have sample data so we can get only December regression line.

22
00:01:42,130 --> 00:01:48,840
We want to use the coefficient of symbol regression lane as an estimate for the population gratingly.

23
00:01:50,870 --> 00:01:53,520
How far off release empl estimate?

24
00:01:53,660 --> 00:01:54,030
That is.

25
00:01:54,100 --> 00:02:00,920
But as you look up and be done, Cap will be from the population coefficients which are be you won't

26
00:02:00,940 --> 00:02:03,050
be done for this.

27
00:02:03,110 --> 00:02:07,130
We will use a quantity called standard error of bedazzling redone.

28
00:02:09,710 --> 00:02:14,760
The standard error of BW and Weedman is calculated using these formulas again.

29
00:02:15,110 --> 00:02:18,110
I'm showing you these formulas and discussing the intuition behind them.

30
00:02:19,070 --> 00:02:23,090
You need not memorize them or know their derivation for practical purposes.

31
00:02:23,840 --> 00:02:29,240
But since these quantities will be part of the output of the model from us or to it, we need to understand

32
00:02:29,240 --> 00:02:29,780
their meaning.

33
00:02:31,750 --> 00:02:32,720
So in this formula.

34
00:02:33,790 --> 00:02:37,990
Delta Sigma squared, sigma squared is the variance of population residuals.

35
00:02:39,670 --> 00:02:46,300
Remember, we discussed residuals, Duell is the value, which is the difference of actual way from

36
00:02:46,300 --> 00:02:48,550
the estimated value of the population.

37
00:02:49,760 --> 00:02:55,600
Sigma squared is the variance of these values for all the data points of the population.

38
00:02:56,920 --> 00:02:59,320
Since population Diggnation Lane is not known.

39
00:02:59,680 --> 00:03:01,390
The Sigma squad is also not known.

40
00:03:02,380 --> 00:03:10,340
We need to estimate it from our sample data to this estimate is given by this formula, which is Oddisee

41
00:03:10,740 --> 00:03:15,650
is equal to under route RSS by and minus to here.

42
00:03:15,850 --> 00:03:18,580
Odyssey is called residual standard error.

43
00:03:19,720 --> 00:03:21,620
Odyssey's is did a little small squares.

44
00:03:22,240 --> 00:03:26,340
We discussed this only last video, which is sum of all the squares.

45
00:03:26,650 --> 00:03:30,000
So for all the data points, we get the errors.

46
00:03:31,140 --> 00:03:32,200
That is the residuals.

47
00:03:32,710 --> 00:03:34,420
We squared them and we add them.

48
00:03:34,450 --> 00:03:36,220
That is the Odyssey's.

49
00:03:37,170 --> 00:03:40,050
We divide Odyssey's by N minus two.

50
00:03:40,320 --> 00:03:41,880
And is the number of observations.

51
00:03:42,030 --> 00:03:47,430
So for our dataset, it is by winders six will divided by five under six, minus two.

52
00:03:48,740 --> 00:03:49,910
Then we find it on the road.

53
00:03:50,180 --> 00:03:52,730
We get there's a dual standard at a.

54
00:03:54,640 --> 00:03:58,600
We'll use this residual standard error as a proxy for Sigma.

55
00:04:00,190 --> 00:04:06,120
Also, notice that if X is more spread out in this formula, B, does it go and be done?

56
00:04:07,780 --> 00:04:15,490
Standard error of bedazzling reduction will be small, which intuitively means that we have more leverage

57
00:04:15,520 --> 00:04:17,770
while estimating the slope in such a case.

58
00:04:19,210 --> 00:04:21,730
Now, what is the practical take away from the S.

59
00:04:21,810 --> 00:04:22,720
E calculations?

60
00:04:23,890 --> 00:04:30,640
Standard error will be used to give us a confidence interval, that is, or linear regression.

61
00:04:31,640 --> 00:04:37,640
There is a 95 percent chance that provably of bidone lies in the interval.

62
00:04:39,060 --> 00:04:44,400
Redoubling gap minus two time standard at it, if we don't do bidone gap.

63
00:04:44,450 --> 00:04:46,460
Plus two things gender identity reduction.

64
00:04:48,400 --> 00:04:52,870
So within this interval, we had Lawwell use this and hired will use this.

65
00:04:54,040 --> 00:04:59,410
We are 95 percent confident that the actual reduction lies in this interval.

66
00:05:01,210 --> 00:05:08,690
Similarly, what we does zero, the 95 percent confidence interval will be the estimates.

67
00:05:08,740 --> 00:05:10,700
We do zero minus two days.

68
00:05:10,710 --> 00:05:11,360
Standard error.

69
00:05:12,100 --> 00:05:12,370
This.

70
00:05:13,330 --> 00:05:14,730
Standard edit of BW.

71
00:05:15,980 --> 00:05:19,780
And the higher value will be be does he look plus two times standard at at all?

72
00:05:19,870 --> 00:05:20,390
What does he do?

73
00:05:22,130 --> 00:05:26,570
To summarize, what we have done here is we had two lines.

74
00:05:26,840 --> 00:05:29,480
One line we got from dissemble regression that we did.

75
00:05:30,530 --> 00:05:36,890
The other one is a hypothetical line, which is the true line between the population of all the points.

76
00:05:38,770 --> 00:05:45,550
We wanted to show whether we can approximate this sample line as the population line.

77
00:05:47,200 --> 00:05:50,240
For that, we found out the confidence interval.

78
00:05:51,540 --> 00:05:55,230
Within which the population coefficient will lay.

79
00:05:56,980 --> 00:06:02,350
So, Debbie, does it all of the population line really between.

80
00:06:03,590 --> 00:06:11,030
These two values that we get from dissemble regression and the slope of the population regression relay.

81
00:06:11,140 --> 00:06:14,150
Between these two values of the sample equation.

82
00:06:16,490 --> 00:06:19,670
And we have a signed up probably of how confident we are.

83
00:06:20,090 --> 00:06:27,740
We are saying that there is a 95 percent chance that the population Diggnation corporations lie within

84
00:06:28,040 --> 00:06:31,340
these intervals, that we are not using sampling.

85
00:06:35,280 --> 00:06:42,150
Another use of standard error is to establish that whether X and Y actually have a relationship or not.

86
00:06:43,800 --> 00:06:47,010
In a linear model, this relationship is given by Biederman.

87
00:06:48,220 --> 00:06:49,090
We are saying that.

88
00:06:49,240 --> 00:06:51,820
Why is be the one times X plus constant?

89
00:06:53,050 --> 00:06:57,270
So if beta one is zero, it means that there is no relationship.

90
00:06:58,900 --> 00:07:02,900
If there is no relationship, then the variable X cannot be used to predict like.

91
00:07:04,040 --> 00:07:08,540
So we need to show that the probability of beta one being zero is negligible.

92
00:07:10,980 --> 00:07:15,510
Let me tell you the contempt first, and after that, I'll show you the way, how you will find it in

93
00:07:15,600 --> 00:07:17,160
every other statistical book.

94
00:07:18,090 --> 00:07:22,080
The concept is this Biederman has some value.

95
00:07:23,070 --> 00:07:27,720
That is the most probable value found by the formula we saw in the previous videos.

96
00:07:29,760 --> 00:07:35,310
Now, if you want to see that we are sufficiently confident that it cannot be zero.

97
00:07:35,880 --> 00:07:37,020
It has two parts to it.

98
00:07:37,560 --> 00:07:43,030
First, most probable value should be fired from zero and taken.

99
00:07:43,710 --> 00:07:49,680
The standard error of Weeda, which is giving us the interval in which the rule lays, should be small.

100
00:07:51,510 --> 00:07:57,430
So basically, we want Z2 not to lie in this whole interval in which be delays.

101
00:07:59,190 --> 00:08:00,220
I hope you get the idea.

102
00:08:01,300 --> 00:08:06,730
Let's do it the proper way, no, this this method is called hypothesis testing.

103
00:08:07,810 --> 00:08:09,820
We will construct two hypotheses.

104
00:08:09,880 --> 00:08:14,770
One is at zero, that is there is no relationship between X and Y.

105
00:08:15,870 --> 00:08:23,490
And its alternative, which is written as it is that there is a literature between X and Y.

106
00:08:24,840 --> 00:08:29,040
So it is Weedman is equal, Luisito.

107
00:08:29,190 --> 00:08:31,140
It is between is not equal, Luisito.

108
00:08:32,500 --> 00:08:34,630
And we want to disprove a digital.

109
00:08:36,760 --> 00:08:38,070
To disprove at zero.

110
00:08:38,520 --> 00:08:45,390
We will calculate something known as B statistic, which is given as BS equal to be done cap minus zero

111
00:08:45,720 --> 00:08:48,300
divided by standard A. B double.

112
00:08:50,190 --> 00:08:55,410
So you can see what this what this is representing here, enumerator.

113
00:08:55,560 --> 00:08:58,800
It says how far Bita one is from zero.

114
00:09:00,020 --> 00:09:04,320
And when you divided by standard at it, you go the distance is how many times?

115
00:09:04,680 --> 00:09:05,330
Standard edit.

116
00:09:06,240 --> 00:09:09,190
So using this formula will go to devalue.

117
00:09:10,610 --> 00:09:13,050
No, we want this devalue to be large.

118
00:09:14,280 --> 00:09:19,230
By the way, it is called P-value, because it is based on B distribution, which is similar to normal

119
00:09:19,230 --> 00:09:19,860
distribution.

120
00:09:21,460 --> 00:09:24,580
Basically, tea distribution is another prohibited distribution.

121
00:09:25,000 --> 00:09:32,440
And if you have P-value, you can get the probability of observing any value equal to absolutely or

122
00:09:32,450 --> 00:09:32,920
logit.

123
00:09:34,720 --> 00:09:39,250
The probability that you will get from this distribution is called P-value.

124
00:09:42,350 --> 00:09:44,120
Now, this P-value has meaning.

125
00:09:45,960 --> 00:09:48,270
We can interpret P-value as follows.

126
00:09:49,400 --> 00:09:55,830
Small value of B will mean that it is highly unlikely that there is no relationship between preterm

127
00:09:55,850 --> 00:09:56,440
response.

128
00:09:58,910 --> 00:10:04,160
Which means that we can reject the null hypothesis and declare that there is a relationship between

129
00:10:04,190 --> 00:10:04,770
X and Y.

130
00:10:07,770 --> 00:10:08,790
Typically, we use.

131
00:10:09,890 --> 00:10:16,050
A value like five percent or one percent as the cutoff value for P, that is, if P is listin point

132
00:10:16,050 --> 00:10:21,810
zero one, then the variable X is significantly impacting Y.

133
00:10:26,240 --> 00:10:28,400
There does look at deserts of of sample.

134
00:10:32,070 --> 00:10:36,890
Or that this is a desert which we have received from the software package that we are using and the

135
00:10:36,980 --> 00:10:44,660
course, there is a separate video where you you will learn how to run this analysis and get this result.

136
00:10:45,890 --> 00:10:51,140
I'm just discussing the result here as an example to all to read that we just covered.

137
00:10:52,580 --> 00:10:59,120
So in the last video, we had just seen the beta one and Bedazzler values this intercept, this debate,

138
00:10:59,120 --> 00:10:59,610
as you know.

139
00:11:00,500 --> 00:11:04,190
And this room, numb with evil, has this slope.

140
00:11:05,140 --> 00:11:08,230
This nine point zero nine is the beta one for this very evil.

141
00:11:10,960 --> 00:11:13,060
The first thing that we covered is the standard error.

142
00:11:13,460 --> 00:11:14,890
What does he do when we done?

143
00:11:16,870 --> 00:11:22,800
The formula that I showed you earlier was used to calculate the standard error of Bedazzler and be done

144
00:11:24,040 --> 00:11:25,840
how the standard elegant me use this.

145
00:11:26,650 --> 00:11:35,020
You can make this statement that for the sample of five, six observations, the beads that you calculated

146
00:11:35,020 --> 00:11:37,360
for minimum error was nine point zero nine.

147
00:11:39,400 --> 00:11:47,740
But you are 95 percent confident that for the global data of how spacing the be done will lay between

148
00:11:48,370 --> 00:11:52,810
nine point zero nine minus two times zero point for one.

149
00:11:54,110 --> 00:11:58,460
Do nine point zero nine plus two times zero point four to.

150
00:12:00,780 --> 00:12:06,930
So this interval is giving you that the estimated B does that you have here.

151
00:12:08,230 --> 00:12:15,940
Can be used to make this statement that deep-Rooted aggression coalitions will lie in the interval given

152
00:12:15,940 --> 00:12:16,910
by the standard error.

153
00:12:18,930 --> 00:12:22,890
The next thing we discussed will be when you and BP will you?

154
00:12:25,300 --> 00:12:26,760
This devaluing p value.

155
00:12:26,920 --> 00:12:29,830
We said that we have one hypothesis.

156
00:12:29,920 --> 00:12:31,820
We're just saying that there is no relationship.

157
00:12:33,930 --> 00:12:37,590
To disprove that hypothesis, we calculated or devalue.

158
00:12:37,680 --> 00:12:39,270
Which was given by this formula.

159
00:12:42,320 --> 00:12:49,400
So to get the value of this be done, we will put the bit of uncapped value divided by the S.

160
00:12:49,490 --> 00:12:55,100
E, so nine point zero nine divided by zero point forty one gives you this devalue.

161
00:12:56,210 --> 00:13:03,010
So corresponding to this devalue, we calculate a P value, which is it as probability of getting a

162
00:13:03,010 --> 00:13:05,930
P value, which is greater than equal to this?

163
00:13:05,960 --> 00:13:07,090
The that we call ridded.

164
00:13:08,120 --> 00:13:10,820
This property value is coming out to be very small.

165
00:13:11,000 --> 00:13:17,330
It's a it's an exponential with a bout of minus 16 to this value is very small.

166
00:13:18,680 --> 00:13:29,510
So if this value is very small, we can see that it is unlikely that there is no relationship, which

167
00:13:29,510 --> 00:13:32,270
means that there is some relationship.

168
00:13:33,960 --> 00:13:43,260
And therefore, we are confident that room number variable is impacting the house pricing and the relationship

169
00:13:43,260 --> 00:13:46,380
between them is given by this coefficient.

170
00:13:48,490 --> 00:13:54,410
So the takeaway from this lecture is we have be done and B, does it all values calculated for the simple.

171
00:13:55,900 --> 00:13:59,740
We calculated a standard error, which is helping us in determining two things.

172
00:13:59,830 --> 00:14:04,870
One is or does the event in which the true value of Beta one and B, does it relate?

173
00:14:05,260 --> 00:14:10,390
And the second thing is whether there is actually a relationship between these predictor.

174
00:14:10,420 --> 00:14:11,740
And the response variables.

175
00:14:12,870 --> 00:14:19,000
To establish that there is a listenership we calculated at P-value using these two things, using a

176
00:14:19,010 --> 00:14:21,020
T value, we calculated a P value.

177
00:14:21,450 --> 00:14:27,840
And if this P value is less than a threshold of one percent or five percent, whichever you like to

178
00:14:27,840 --> 00:14:30,570
use, then we see that there is a relationship.

179
00:14:31,990 --> 00:14:37,690
So far, our model, there is a relationship of hose pricing with room them, and that relationship

180
00:14:37,780 --> 00:14:40,220
is this between which is main point you don't make.
