abdullah's picture
Add files using upload-large-folder tool
8de7e09 verified
raw
history blame
36.6 kB
1
00:00:11,500 --> 00:00:17,380
Last time, we talked about chi-square tests. And
2
00:00:17,380 --> 00:00:21,580
we mentioned that there are two objectives in this
3
00:00:21,580 --> 00:00:25,220
chapter. The first one is when to use chi-square
4
00:00:25,220 --> 00:00:28,630
tests for contingency tables. And the other
5
00:00:28,630 --> 00:00:31,070
objective is how to use chi-square tests for
6
00:00:31,070 --> 00:00:35,410
contingency tables. And we did one chi-square test
7
00:00:35,410 --> 00:00:42,050
for the difference between two proportions. In the
8
00:00:42,050 --> 00:00:44,630
null hypothesis, the two proportions are equal. I
9
00:00:44,630 --> 00:00:47,970
mean, proportion for population 1 equals
10
00:00:47,970 --> 00:00:52,970
population proportion 2 against the alternative
11
00:00:52,970 --> 00:00:58,470
here is two-sided test. Pi 1 does not equal pi 2.
12
00:00:59,310 --> 00:01:04,210
In this case, we can use either this statistic. So
13
00:01:04,210 --> 00:01:05,150
you may
14
00:01:07,680 --> 00:01:15,520
Z statistic, which is b1 minus b2 minus y1 minus
15
00:01:15,520 --> 00:01:21,840
y2 divided by b
16
00:01:21,840 --> 00:01:27,500
dash times 1 minus b dash multiplied by 1 over n1
17
00:01:27,500 --> 00:01:31,200
plus 1 over n2. This quantity under the square
18
00:01:31,200 --> 00:01:33,080
root, where b dash
19
00:01:42,180 --> 00:01:48,580
Or proportionally, where P dash equals X1 plus X2
20
00:01:48,580 --> 00:01:55,560
divided by N1 plus N2. Or,
21
00:01:58,700 --> 00:02:00,720
in this chapter, we are going to use chi-square
22
00:02:00,720 --> 00:02:04,520
statistic, which is given by this equation. Chi
23
00:02:04,520 --> 00:02:09,620
-square statistic is just sum of observed
24
00:02:09,620 --> 00:02:10,940
frequency, FO.
25
00:02:15,530 --> 00:02:20,070
minus expected frequency squared divided by
26
00:02:20,070 --> 00:02:22,490
expected frequency for all cells.
27
00:02:25,210 --> 00:02:29,070
Chi squared, this statistic is given by this
28
00:02:29,070 --> 00:02:34,190
equation. If there are two by two rows and
29
00:02:34,190 --> 00:02:36,290
columns, I mean there are two rows and two
30
00:02:36,290 --> 00:02:40,270
columns. So in this case, my table is two by two.
31
00:02:42,120 --> 00:02:44,360
In this case, you have only one degree of freedom.
32
00:02:44,640 --> 00:02:50,440
Always degrees of freedom equals number of rows
33
00:02:50,440 --> 00:03:00,320
minus one multiplied by number of columns minus
34
00:03:00,320 --> 00:03:06,140
one. So for two by two tables, there are two rows
35
00:03:06,140 --> 00:03:11,560
and two columns, so two minus one. times 2 minus
36
00:03:11,560 --> 00:03:15,420
1, so your degrees of freedom in this case is 1.
37
00:03:16,440 --> 00:03:19,320
Here the assumption is we assume that the expected
38
00:03:19,320 --> 00:03:22,940
frequency is at least 5, in order to use Chi
39
00:03:22,940 --> 00:03:27,200
-square statistic. Chi-square is always positive,
40
00:03:27,680 --> 00:03:32,260
I mean, Chi-square value is always greater than 0.
41
00:03:34,040 --> 00:03:38,890
It's one TLTS to the right one. We reject F0 if
42
00:03:38,890 --> 00:03:42,430
your chi-square statistic falls in the rejection
43
00:03:42,430 --> 00:03:45,850
region. That means we reject the null hypothesis
44
00:03:45,850 --> 00:03:49,470
if chi-square statistic greater than chi-square
45
00:03:49,470 --> 00:03:53,130
alpha. Alpha can be determined by using chi-square
46
00:03:53,130 --> 00:03:56,490
table. So we reject in this case F0, otherwise,
47
00:03:56,890 --> 00:04:02,050
sorry, we don't reject F0. So again, if the value
48
00:04:02,050 --> 00:04:05,350
of chi-square statistic falls in this rejection
49
00:04:05,350 --> 00:04:10,280
region, the yellow one, then we reject. Otherwise,
50
00:04:11,100 --> 00:04:13,900
if this value, I mean if the value of the
51
00:04:13,900 --> 00:04:17,060
statistic falls in non-rejection region, we don't
52
00:04:17,060 --> 00:04:21,680
reject the null hypothesis. So the same concept as
53
00:04:21,680 --> 00:04:27,680
we did in the previous chapters. If we go back to
54
00:04:27,680 --> 00:04:32,060
the previous example we had discussed before, when
55
00:04:32,060 --> 00:04:36,620
we are testing about gender and left and right
56
00:04:36,620 --> 00:04:41,340
handers, So hand preference either left or right.
57
00:04:42,960 --> 00:04:49,320
And the question is test to see whether hand
58
00:04:49,320 --> 00:04:53,100
preference and gender are related or not. In this
59
00:04:53,100 --> 00:04:56,960
case, your null hypothesis could be written as
60
00:04:56,960 --> 00:05:00,620
either X0.
61
00:05:04,220 --> 00:05:07,160
So the proportion of left-handers for female
62
00:05:07,160 --> 00:05:12,260
equals the proportion of males left-handers. So by
63
00:05:12,260 --> 00:05:16,600
one equals by two or H zero later we'll see that
64
00:05:16,600 --> 00:05:22,680
the two variables of interest are independent.
65
00:05:32,810 --> 00:05:37,830
Now, your B dash is
66
00:05:37,830 --> 00:05:42,250
given by X1 plus X2 divided by N1 plus N2. X1 is
67
00:05:42,250 --> 00:05:51,930
12, this 12, plus 24 divided by 300. That will
68
00:05:51,930 --> 00:05:57,310
give 12%. So let me just write this notation, B
69
00:05:57,310 --> 00:05:57,710
dash.
70
00:06:05,560 --> 00:06:13,740
equals 36 by 300, so that's 12%. So the expected
71
00:06:13,740 --> 00:06:19,500
frequency in this case for female, 0.12 times 120,
72
00:06:19,680 --> 00:06:22,140
because there are 120 females in the data you
73
00:06:22,140 --> 00:06:25,520
have, so that will give 14.4. So the expected
74
00:06:25,520 --> 00:06:30,680
frequency is 0.12 times 180, 120, I'm sorry,
75
00:06:34,810 --> 00:06:39,590
That will give 14.4. Similarly, for male to be
76
00:06:39,590 --> 00:06:43,390
left-handed is 0.12 times number of females in the
77
00:06:43,390 --> 00:06:46,690
sample, which is 180, and that will give 21.6.
78
00:06:48,670 --> 00:06:53,190
Now, since you compute the expected for the first
79
00:06:53,190 --> 00:06:57,590
cell, the second one direct is just the complement
80
00:06:57,590 --> 00:07:03,020
120. 120 is sample size for the Rome. I mean
81
00:07:03,020 --> 00:07:12,200
female total 120 minus 14.4 will give 105.6. Or 0
82
00:07:12,200 --> 00:07:18,050
.88 times 120 will give the same value. Here, the
83
00:07:18,050 --> 00:07:21,730
expected is 21.6, so the compliment is the, I'm
84
00:07:21,730 --> 00:07:25,130
sorry, the expected is just the compliment, which
85
00:07:25,130 --> 00:07:32,010
is 180 minus 21.6 will give 158.4. Or 0.88 is the
86
00:07:32,010 --> 00:07:35,090
compliment of that one multiplied by 180 will give
87
00:07:35,090 --> 00:07:39,070
the same value. So that's the one we had discussed
88
00:07:39,070 --> 00:07:39,670
before.
89
00:07:42,410 --> 00:07:46,550
On this result, you can determine the value of chi
90
00:07:46,550 --> 00:07:50,750
-square statistic by using this equation. Sum of F
91
00:07:50,750 --> 00:07:53,810
observed minus F expected squared divided by F
92
00:07:53,810 --> 00:07:57,450
expected for each cell. You have to compute the
93
00:07:57,450 --> 00:08:00,870
value of chi-square for each cell. In this case,
94
00:08:01,070 --> 00:08:04,250
the simplest case is just 2 by 2 table. So 12
95
00:08:04,250 --> 00:08:09,980
minus 14.4 squared divided by 14.4. Plus the
96
00:08:09,980 --> 00:08:15,720
second one 108 minus 105 squared divided by 105 up
97
00:08:15,720 --> 00:08:19,780
to the last one, you will get this result. Now my
98
00:08:19,780 --> 00:08:21,900
chi-square value is 0.7576.
99
00:08:24,240 --> 00:08:28,140
And in this case, if chi-square value is very
100
00:08:28,140 --> 00:08:31,180
small, I mean it's close to zero, then we don't
101
00:08:31,180 --> 00:08:34,140
reject the null hypothesis. Because the smallest
102
00:08:34,140 --> 00:08:37,500
value of chi-square is zero, and zero happens only
103
00:08:37,500 --> 00:08:43,580
if f observed is close to f expected. So here if
104
00:08:43,580 --> 00:08:46,920
you look carefully for the observed and expected
105
00:08:46,920 --> 00:08:50,520
frequencies, you can tell if you can reject or
106
00:08:50,520 --> 00:08:53,700
don't reject the number. Now the difference
107
00:08:53,700 --> 00:08:58,960
between these values looks small, so that's lead
108
00:08:58,960 --> 00:09:05,110
to small chi-square. So without doing the critical
109
00:09:05,110 --> 00:09:08,210
value, computer critical value, you can determine
110
00:09:08,210 --> 00:09:11,890
that we don't reject the null hypothesis. Because
111
00:09:11,890 --> 00:09:16,070
your chi-square value is very small. So we don't
112
00:09:16,070 --> 00:09:18,670
reject the null hypothesis. Or if you look
113
00:09:18,670 --> 00:09:22,790
carefully at the table, for the table we have
114
00:09:22,790 --> 00:09:26,410
here, for chi-square table. By the way, the
115
00:09:26,410 --> 00:09:31,480
smallest value of chi-square is 1.3. under 1
116
00:09:31,480 --> 00:09:36,180
degrees of freedom. So the smallest value 1.32. So
117
00:09:36,180 --> 00:09:39,360
if your chi-square value is greater than 1, it
118
00:09:39,360 --> 00:09:41,920
means maybe you reject or don't reject. It depends
119
00:09:41,920 --> 00:09:45,920
on v value and alpha you have or degrees of
120
00:09:45,920 --> 00:09:50,280
freedom. But in the worst scenario, if your chi
121
00:09:50,280 --> 00:09:53,780
-square is smaller than this value, it means you
122
00:09:53,780 --> 00:09:57,600
don't reject the null hypothesis. So generally
123
00:09:57,600 --> 00:10:02,120
speaking, if Chi-square is statistical. It's
124
00:10:02,120 --> 00:10:06,420
smaller than 1.32. 1.32 is a very small value.
125
00:10:06,940 --> 00:10:15,560
Then we don't reject. Then we don't reject x0.
126
00:10:15,780 --> 00:10:24,220
That's always, always true. Regardless of degrees
127
00:10:24,220 --> 00:10:31,050
of freedom and alpha. My chi-square is close to
128
00:10:31,050 --> 00:10:35,710
zero, or smaller than 1.32, because the minimum
129
00:10:35,710 --> 00:10:40,990
value of critical value is 1.32. Imagine that we
130
00:10:40,990 --> 00:10:46,050
are talking about alpha is 5%. So alpha is 5, so
131
00:10:46,050 --> 00:10:48,750
your critical value, the smallest one for 1
132
00:10:48,750 --> 00:10:53,850
degrees of freedom, is 3.84. So that's my
133
00:10:53,850 --> 00:10:55,490
smallest, if alpha
134
00:11:03,740 --> 00:11:08,680
Last time we mentioned that this value is just 1
135
00:11:08,680 --> 00:11:17,760
.96 squared. And that's only true, only true for 2
136
00:11:17,760 --> 00:11:24,180
by 2 table. That means this square is just Chi
137
00:11:24,180 --> 00:11:29,470
square 1. For this reason, we can test by one
138
00:11:29,470 --> 00:11:33,330
equal by two, by two methods, either this
139
00:11:33,330 --> 00:11:37,750
statistic or chi-square statistic. Both of them
140
00:11:37,750 --> 00:11:41,970
will give the same result. So let's go back to the
141
00:11:41,970 --> 00:11:49,670
question we have. My chi-square value is 0.77576.
142
00:11:52,160 --> 00:11:56,940
So that's your chi-square statistic. Again,
143
00:11:57,500 --> 00:12:00,240
degrees of freedom 1 to chi-square, the critical
144
00:12:00,240 --> 00:12:08,500
value is 3.841. So my decision is we don't reject
145
00:12:08,500 --> 00:12:11,780
the null hypothesis. My conclusion is there is not
146
00:12:11,780 --> 00:12:14,380
sufficient evidence that two proportions are
147
00:12:14,380 --> 00:12:17,480
different. So you don't have sufficient evidence
148
00:12:17,480 --> 00:12:21,900
in order to support that the two proportions are
149
00:12:21,900 --> 00:12:27,720
different at 5% level of significance. We stopped
150
00:12:27,720 --> 00:12:32,700
last time at this point. Now suppose we are
151
00:12:32,700 --> 00:12:36,670
testing The difference among more than two
152
00:12:36,670 --> 00:12:42,930
proportions. The same steps, we have to extend in
153
00:12:42,930 --> 00:12:47,830
this case chi-square. Your null hypothesis, by one
154
00:12:47,830 --> 00:12:50,990
equal by two, all the way up to by C. So in this
155
00:12:50,990 --> 00:13:00,110
case, there are C columns. C columns and
156
00:13:00,110 --> 00:13:05,420
two rows. So number of columns equals C, and there
157
00:13:05,420 --> 00:13:10,520
are only two rows. So pi 1 equals pi 2, all the
158
00:13:10,520 --> 00:13:13,840
way up to pi C. So null hypothesis for the columns
159
00:13:13,840 --> 00:13:17,040
we have. There are C columns. Again, it's the
160
00:13:17,040 --> 00:13:19,840
alternative, not all of the pi J are equal, and J
161
00:13:19,840 --> 00:13:23,840
equals 1 up to C. Now, the only difference here,
162
00:13:26,520 --> 00:13:27,500
the degrees of freedom.
163
00:13:31,370 --> 00:13:32,850
For 2 by c table,
164
00:13:35,710 --> 00:13:42,010
2 by c, degrees of freedom equals number
165
00:13:42,010 --> 00:13:45,890
of rows minus 1. There are two rows, so 2 minus 1
166
00:13:45,890 --> 00:13:50,810
times number of columns minus 1. 2 minus 1 is 1, c
167
00:13:50,810 --> 00:13:54,610
minus 1, 1 times c minus 1, c minus 1. So your
168
00:13:54,610 --> 00:13:57,130
degrees of freedom in this case is c minus 1.
169
00:14:00,070 --> 00:14:03,190
So that's the only difference. For two by two
170
00:14:03,190 --> 00:14:07,130
table, degrees of freedom is just one. If there
171
00:14:07,130 --> 00:14:10,670
are C columns and we have the same number of rows,
172
00:14:11,450 --> 00:14:14,810
degrees of freedom is C minus one. And we have the
173
00:14:14,810 --> 00:14:19,190
same chi squared statistic, the same equation I
174
00:14:19,190 --> 00:14:23,890
mean. And we have to extend also the overall
175
00:14:23,890 --> 00:14:27,330
proportion instead of x1 plus x2 divided by n1
176
00:14:27,330 --> 00:14:32,610
plus n2. It becomes x1 plus x2 plus x3 all the way
177
00:14:32,610 --> 00:14:38,330
up to xc because there are c columns divided by n1
178
00:14:38,330 --> 00:14:41,910
plus n2 all the way up to nc. So that's x over n.
179
00:14:43,540 --> 00:14:48,400
So similarly we can reject the null hypothesis if
180
00:14:48,400 --> 00:14:52,260
the value of chi-square statistic lies or falls in
181
00:14:52,260 --> 00:14:54,160
the rejection region.
182
00:14:58,120 --> 00:15:01,980
Other type of chi-square test is called chi-square
183
00:15:01,980 --> 00:15:07,380
test of independence. Generally speaking, most of
184
00:15:07,380 --> 00:15:10,440
the time there are more than two columns or more
185
00:15:10,440 --> 00:15:16,490
than two rows. Now, suppose we have contingency
186
00:15:16,490 --> 00:15:22,370
table that has R rows and C columns. And we are
187
00:15:22,370 --> 00:15:26,990
interested to test to see whether the two
188
00:15:26,990 --> 00:15:31,390
categorical variables are independent. That means
189
00:15:31,390 --> 00:15:35,600
there is no relationship between them. Against the
190
00:15:35,600 --> 00:15:38,80
223
00:18:05,940 --> 00:18:11,560
tables with R rows and C columns. So we have the
224
00:18:11,560 --> 00:18:15,660
case R by C. So that's in general, there are R
225
00:18:15,660 --> 00:18:23,060
rows and C columns. And the question is this C, if
226
00:18:23,060 --> 00:18:27,480
the two variables are independent or not. So in
227
00:18:27,480 --> 00:18:30,700
this case, you cannot use this statistic. So one
228
00:18:30,700 --> 00:18:34,320
more time, this statistic is valid only for two by
229
00:18:34,320 --> 00:18:38,020
two tables. So that means we can use z or chi
230
00:18:38,020 --> 00:18:41,200
-square to test if there is no difference between
231
00:18:41,200 --> 00:18:43,960
two population proportions. But for more than
232
00:18:43,960 --> 00:18:46,700
that, you have to use chi-square.
233
00:18:49,950 --> 00:18:53,310
Now still we have the same equation, Chi-square
234
00:18:53,310 --> 00:18:57,870
statistic is just sum F observed minus F expected
235
00:18:57,870 --> 00:19:00,690
quantity squared divided by F expected.
236
00:19:03,490 --> 00:19:07,550
In this case, Chi-square statistic for R by C case
237
00:19:07,550 --> 00:19:15,430
has degrees of freedom R minus 1 multiplied by C
238
00:19:15,430 --> 00:19:18,570
minus 1. In this case, each cell in the
239
00:19:18,570 --> 00:19:21,230
contingency table has expected frequency at least
240
00:19:21,230 --> 00:19:26,910
one instead of five. Now let's see how can we
241
00:19:26,910 --> 00:19:31,690
compute the expected cell frequency for each cell.
242
00:19:32,950 --> 00:19:37,530
The expected frequency is given by row total
243
00:19:37,530 --> 00:19:42,950
multiplied by colon total divided by n. So that's
244
00:19:42,950 --> 00:19:50,700
my new equation to determine I've expected it. So
245
00:19:50,700 --> 00:19:56,440
the expected value for each cell is given by Rho
246
00:19:56,440 --> 00:20:03,380
total multiplied by Kono, total divided by N.
247
00:20:05,160 --> 00:20:09,540
Also, this equation is true for the previous
248
00:20:09,540 --> 00:20:15,560
example. If you go back a little bit here, now the
249
00:20:16,650 --> 00:20:21,650
Expected for this cell was 40.4. Now let's see how
250
00:20:21,650 --> 00:20:25,470
can we compute the same value by using this
251
00:20:25,470 --> 00:20:30,250
equation. So it's equal to row total 120
252
00:20:30,250 --> 00:20:40,310
multiplied by column total 36 divided by 300.
253
00:20:43,580 --> 00:20:46,500
Now before we compute this value by using B dash
254
00:20:46,500 --> 00:20:50,900
first, 300 divided by, I'm sorry, 36 divided by
255
00:20:50,900 --> 00:20:58,520
300. So that's your B dash. Then we multiply this
256
00:20:58,520 --> 00:21:03,540
B dash by N, and this is your N. So it's similar
257
00:21:03,540 --> 00:21:08,540
equation. So either you use row total multiplied
258
00:21:08,540 --> 00:21:14,060
by column total. then divide by overall sample
259
00:21:14,060 --> 00:21:18,880
size you will get the same result by using the
260
00:21:18,880 --> 00:21:25,520
overall proportion 12% times 120 so each one will
261
00:21:25,520 --> 00:21:29,860
give the same answer so from now we are going to
262
00:21:29,860 --> 00:21:33,900
use this equation in order to compute the expected
263
00:21:33,900 --> 00:21:37,960
frequency for each cell so again expected
264
00:21:37,960 --> 00:21:42,920
frequency is rho total times Column total divided
265
00:21:42,920 --> 00:21:48,620
by N, N is the sample size. So row total it means
266
00:21:48,620 --> 00:21:52,220
sum of all frequencies in the row. Similarly
267
00:21:52,220 --> 00:21:56,160
column total is the sum of all frequencies in the
268
00:21:56,160 --> 00:22:00,180
column and N is over all sample size.
269
00:22:03,030 --> 00:22:06,630
Again, we reject the null hypothesis if your chi
270
00:22:06,630 --> 00:22:10,430
-square statistic greater than chi-square alpha.
271
00:22:10,590 --> 00:22:13,370
Otherwise, you don't reject it. And keep in mind,
272
00:22:14,270 --> 00:22:18,390
chi-square statistic has degrees of freedom R
273
00:22:18,390 --> 00:22:23,730
minus 1 times C minus 1. That's all for chi-square
274
00:22:23,730 --> 00:22:27,590
as test of independence. Any question?
275
00:22:31,220 --> 00:22:36,300
Here there is an example for applying chi-square
276
00:22:36,300 --> 00:22:42,200
test of independence. Meal plan selected
277
00:22:42,200 --> 00:22:46,700
by 200 students is shown in this table. So there
278
00:22:46,700 --> 00:22:50,960
are two variables of interest. The first one is
279
00:22:50,960 --> 00:22:56,230
number of meals per week. And there are three
280
00:22:56,230 --> 00:23:00,550
types of number of meals, either 20 meals per
281
00:23:00,550 --> 00:23:07,870
week, or 10 meals per week, or none. So that's, so
282
00:23:07,870 --> 00:23:12,150
number of meals is classified into three groups.
283
00:23:13,210 --> 00:23:17,650
So three columns, 20 per week, 10 per week, or
284
00:23:17,650 --> 00:23:23,270
none. Class standing, students are classified into
285
00:23:23,270 --> 00:23:28,860
four levels. A freshman, it means students like
286
00:23:28,860 --> 00:23:33,620
you, first year. Sophomore, it means second year.
287
00:23:34,440 --> 00:23:38,400
Junior, third level. Senior, fourth level. So that
288
00:23:38,400 --> 00:23:42,100
means first, second, third, and fourth level. And
289
00:23:42,100 --> 00:23:46,140
we have this number, these numbers for, I mean,
290
00:23:47,040 --> 00:23:53,660
there are 24 A freshman who have meals for 20 per
291
00:23:53,660 --> 00:23:59,880
week. So there are 24 freshmen have 20 meals per
292
00:23:59,880 --> 00:24:04,160
week. 22 sophomores, the same, 10 for junior and
293
00:24:04,160 --> 00:24:10,220
14 for senior. And the question is just to see if
294
00:24:10,220 --> 00:24:13,740
number of meals per week is independent of class
295
00:24:13,740 --> 00:24:17,270
standing. to see if there is a relationship
296
00:24:17,270 --> 00:24:21,890
between these two variables. In this case, there
297
00:24:21,890 --> 00:24:26,850
are four rows because the class standing is
298
00:24:26,850 --> 00:24:29,190
classified into four groups. So there are four
299
00:24:29,190 --> 00:24:34,230
rows and three columns. So this table actually is
300
00:24:34,230 --> 00:24:40,200
four by three. And there are twelve cells in this
301
00:24:40,200 --> 00:24:46,660
case. Now it takes time to compute the expected
302
00:24:46,660 --> 00:24:49,760
frequencies because in this case we have to
303
00:24:49,760 --> 00:24:55,120
compute the expected frequency for each cell. And
304
00:24:55,120 --> 00:25:01,320
we are going to use this formula for only six of
305
00:25:01,320 --> 00:25:06,260
them. I mean, we can apply this formula for only
306
00:25:06,260 --> 00:25:09,880
six of them. And the others can be computed by the
307
00:25:09,880 --> 00:25:14,300
complement by using either column total or row
308
00:25:14,300 --> 00:25:19,940
total. So because degrees of freedom is six, that
309
00:25:19,940 --> 00:25:23,880
means you may use this rule six times only. The
310
00:25:23,880 --> 00:25:28,420
others can be computed by using the complement. So
311
00:25:28,420 --> 00:25:34,070
here again, the hypothesis to be tested is, Mean
312
00:25:34,070 --> 00:25:36,550
plan and class standing are independent, that
313
00:25:36,550 --> 00:25:38,670
means there is no relationship between them.
314
00:25:39,150 --> 00:25:41,650
Against alternative hypothesis, mean plan and
315
00:25:41,650 --> 00:25:44,630
class standing are dependent, that means there
316
00:25:44,630 --> 00:25:49,950
exists significant relationship between them. Now
317
00:25:49,950 --> 00:25:54,390
let's see how can we compute the expected cell,
318
00:25:55,990 --> 00:26:00,470
the expected frequency for each cell. For example,
319
00:26:02,250 --> 00:26:07,790
The first observed frequency is 24. Now the
320
00:26:07,790 --> 00:26:15,990
expected should be 70 times 70 divided by 200. So
321
00:26:15,990 --> 00:26:25,050
for cell 11, the first cell. If expected, we can
322
00:26:25,050 --> 00:26:32,450
use this notation, 11. Means first row. First
323
00:26:32,450 --> 00:26:40,110
column. That should be 70. It is 70. Multiplied by
324
00:26:40,110 --> 00:26:43,990
column totals. Again, in this case, 70. Multiplied
325
00:26:43,990 --> 00:26:47,270
by 200. That will give 24.5.
326
00:26:50,150 --> 00:26:53,730
Similarly, for the second cell, for 32.
327
00:26:56,350 --> 00:27:00,090
70 times 88 divided by 200.
328
00:27:02,820 --> 00:27:12,620
So for F22, again it's 70 times 88 divided by 200,
329
00:27:12,800 --> 00:27:22,060
that will get 30.8. So 70 times 88, that will give
330
00:27:22,060 --> 00:27:32,780
30.8. F21, rule two first, one third. rho 1 second
331
00:27:32,780 --> 00:27:37,600
one the third one now either you can use the same
332
00:27:37,600 --> 00:27:44,320
equation which is 70 times 42 so you can use 70
333
00:27:44,320 --> 00:27:54,360
times 42 divided by 200 that will give 14.7 or
334
00:27:54,360 --> 00:27:59,000
it's just the complement which is 70 minus
335
00:28:03,390 --> 00:28:14,510
24.5 plus 30.8. So either use 70 multiplied by 40
336
00:28:14,510 --> 00:28:19,390
divided by 200 or just the complement, 70 minus.
337
00:28:20,800 --> 00:28:28,400
24.5 plus 30.8 will give the same value. So I just
338
00:28:28,400 --> 00:28:32,740
compute the expected cell for 1 and 2, and the
339
00:28:32,740 --> 00:28:36,120
third one is just the complement. Similarly, for
340
00:28:36,120 --> 00:28:42,560
the second row, I mean cell 21, then 22, and 23.
341
00:28:43,680 --> 00:28:47,940
By using the same method, he will get these two
342
00:28:47,940 --> 00:28:51,880
values, and the other one is the complement, which
343
00:28:51,880 --> 00:28:54,880
is 60 minus these, the sum of these two values,
344
00:28:55,300 --> 00:28:55,960
will give 12.
345
00:28:58,720 --> 00:29:01,920
Similarly, for the third cell, I'm sorry, the
346
00:29:01,920 --> 00:29:07,460
third row, for this value, For 10, it's 30 times
347
00:29:07,460 --> 00:29:12,660
70 divided by 200 will give this result. And the
348
00:29:12,660 --> 00:29:16,060
other one is just 30 multiplied by 88 divided by
349
00:29:16,060 --> 00:29:20,200
200. The other one is just the complement, 30
350
00:29:20,200 --> 00:29:25,180
minus the sum of these. Now, for the last column,
351
00:29:26,660 --> 00:29:35,220
either 70 multiplied by 70 divided by 200, or 70
352
00:29:35,220 --> 00:29:41,780
this 70 minus the sum of these. 70 this one equals
353
00:29:41,780 --> 00:29:51,740
70 minus the sum of 24 plus 21 plus 10. That will
354
00:29:51,740 --> 00:30:01,120
give 14. Now for the other expected cell, 88.
355
00:30:02,370 --> 00:30:05,530
minus the sum of these three expected frequencies.
356
00:30:07,290 --> 00:30:12,810
Now for the last one, last one is either by 42
357
00:30:12,810 --> 00:30:17,770
minus the sum of these three, or 40 minus the sum
358
00:30:17,770 --> 00:30:20,090
of 14 plus 6, 17.6.
359
00:30:22,810 --> 00:30:27,940
Or 40 multiplied by 42 divided by 400. So let's
360
00:30:27,940 --> 00:30:35,180
say we use that formula six times. For this
361
00:30:35,180 --> 00:30:39,100
reason, degrees of freedom is six. The other six
362
00:30:39,100 --> 00:30:46,480
are computed by the complement as we mentioned. So
363
00:30:46,480 --> 00:30:50,240
these are the expected frequencies. It takes time
364
00:30:50,240 --> 00:30:56,010
to compute these. But if you have only two by two
365
00:30:56,010 --> 00:31:01,170
table, it's easier. Now based on that, we can
366
00:31:01,170 --> 00:31:07,430
compute chi-square statistic value by using this
367
00:31:07,430 --> 00:31:12,390
equation for each cell. I mean, the first one, if
368
00:31:12,390 --> 00:31:14,370
you go back a little bit to the previous table,
369
00:31:15,150 --> 00:31:18,130
here, in order to compute chi-square,
370
00:31:22,640 --> 00:31:27,760
value, we have to use this equation, pi squared,
371
00:31:28,860 --> 00:31:36,080
sum F observed minus F expected squared, divided
372
00:31:36,080 --> 00:31:41,980
by F expected for all C's. So the first one is 24
373
00:31:41,980 --> 00:31:44,780
minus squared,
374
00:31:46,560 --> 00:31:55,350
24 plus. The second cell is 32 squared
375
00:31:55,350 --> 00:31:58,990
plus
376
00:31:58,990 --> 00:32:02,930
all the way up to the last cell, which is 10.
377
00:32:11,090 --> 00:32:14,430
So it takes time. But again, for two by two, it's
378
00:32:14,430 --> 00:32:18,890
straightforward. Anyway, now if you compare the
379
00:32:18,890 --> 00:32:23,650
expected and observed cells, you can have an idea
380
00:32:23,650 --> 00:32:25,650
either to reject or fail to reject without
381
00:32:25,650 --> 00:32:31,470
computing the value itself. Now, 24, 24.5. The
382
00:32:31,470 --> 00:32:32,430
difference is small.
383
00:32:35,730 --> 00:32:39,070
for about 7 and so on. So the difference between
384
00:32:39,070 --> 00:32:44,450
observed and expected looks small. In this case,
385
00:32:44,590 --> 00:32:50,530
chi-square value is close to zero. So it's 709.
386
00:32:51,190 --> 00:32:55,370
Now, without looking at the table we have, we have
387
00:32:55,370 --> 00:33:02,710
to don't reject. So we don't reject Because as we
388
00:33:02,710 --> 00:33:05,450
mentioned, the minimum k squared value is 1132.
389
00:33:06,350 --> 00:33:09,670
That's for one degrees of freedom and the alpha is
390
00:33:09,670 --> 00:33:14,390
25%. So
391
00:33:14,390 --> 00:33:19,250
I expect my decision is don't reject the null
392
00:33:19,250 --> 00:33:24,530
hypothesis. Now by looking at k squared 5% and
393
00:33:24,530 --> 00:33:28,870
degrees of freedom 6 by using k squared theorem.
394
00:33:30,200 --> 00:33:36,260
Now degrees of freedom 6. Now the minimum value of
395
00:33:36,260 --> 00:33:40,520
Chi-square is 7.84. I mean critical value. But
396
00:33:40,520 --> 00:33:48,290
under 5% is 12.59. So this value is 12.59. So
397
00:33:48,290 --> 00:33:54,470
critical value is 12.59. So my rejection region is
398
00:33:54,470 --> 00:33:59,890
above this value. Now, my chi-square value falls
399
00:33:59,890 --> 00:34:06,250
in the non-rejection regions. It's very small
400
00:34:06,250 --> 00:34:13,850
value. So chi-square statistic is 0.709.
401
00:34:14,230 --> 00:34:20,620
It's much smaller. Not even smaller than π²α, it's
402
00:34:20,620 --> 00:34:23,580
much smaller than this value, so it means we don't
403
00:34:23,580 --> 00:34:26,440
have sufficient evidence to support the
404
00:34:26,440 --> 00:34:32,010
alternative hypothesis. So my decision is, don't
405
00:34:32,010 --> 00:34:36,350
reject the null hypothesis. So conclusion, there
406
00:34:36,350
445
00:38:19,540 --> 00:38:25,080
Z statistic or Chi-squared. So Z or Chi can be
446
00:38:25,080 --> 00:38:28,920
used for testing difference between two population
447
00:38:28,920 --> 00:38:34,360
proportions. And again, chi-square can be extended
448
00:38:34,360 --> 00:38:40,140
to use for more than two. So in this case, the
449
00:38:40,140 --> 00:38:43,220
correct answer is C, because we can use either Z
450
00:38:43,220 --> 00:38:52,090
or chi-square test. Next, in testing hypothesis
451
00:38:52,090 --> 00:38:58,350
using chi-square test. The theoretical frequencies
452
00:38:58,350 --> 00:39:03,190
are based on the null hypothesis, alternative, normal
453
00:39:03,190 --> 00:39:06,490
distribution, none of the above. Always when we
454
00:39:06,490 --> 00:39:10,450
are using chi-square test, we assume the null is
455
00:39:10,450 --> 00:39:14,630
true. So the theoretical frequencies are based on
456
00:39:14,630 --> 00:39:20,060
the null hypothesis. So always any statistic can
457
00:39:20,060 --> 00:39:25,300
be computed if we assume x0 is correct. So the
458
00:39:25,300 --> 00:39:26,400
correct answer is A.
459
00:39:34,060 --> 00:39:37,040
Let's look at table 11-2.
460
00:39:44,280 --> 00:39:49,000
Many companies use well-known celebrities as
461
00:39:49,000 --> 00:39:54,420
spokespersons in their TV advertisements. A study
462
00:39:54,420 --> 00:39:57,760
was conducted to determine whether brand awareness
463
00:39:57,760 --> 00:40:02,140
of female TV viewers and the gender of the
464
00:40:02,140 --> 00:40:05,860
spokesperson are independent. So there are two
465
00:40:05,860 --> 00:40:09,820
variables, whether a brand awareness of female TV
466
00:40:09,820 --> 00:40:13,740
and gender of the spokesperson are independent.
467
00:40:14,820 --> 00:40:19,540
Each and a sample of 300 female TV viewers was
468
00:40:19,540 --> 00:40:24,000
asked to identify a product advertised by a
469
00:40:24,000 --> 00:40:27,000
celebrity spokesperson, the gender of the
470
00:40:27,000 --> 00:40:32,280
spokesperson, and whether or not the viewer could
471
00:40:32,280 --> 00:40:36,460
identify the product was recorded. The number in
472
00:40:36,460 --> 00:40:40,080
each category are given below. Now, the questions
473
00:40:40,080 --> 00:40:45,520
are, number one, he asked about the calculated
474
00:40:45,520 --> 00:40:49,120
this statistic is. We have to find Chi-square
475
00:40:49,120 --> 00:40:54,020
statistic. It's two by two tables, easy one. So,
476
00:40:54,460 --> 00:40:59,460
for example, to find the F expected is,
477
00:41:00,420 --> 00:41:13,130
rho total is one over two. And one line here. And
478
00:41:13,130 --> 00:41:13,810
this 150.
479
00:41:16,430 --> 00:41:22,510
And also 150. So the expected frequency for the
480
00:41:22,510 --> 00:41:31,010
first one is 102 times 150 divided by 300.
481
00:41:35,680 --> 00:41:39,640
So the answer is 51.
482
00:41:42,880 --> 00:41:51,560
So the first expected is 51. The other one is just
483
00:41:51,560 --> 00:41:54,360
102 minus 51 is also 51.
484
00:41:57,320 --> 00:42:01,020
Now here is 99.
485
00:42:09,080 --> 00:42:15,180
So the second
486
00:42:15,180 --> 00:42:18,800
one are the expected frequencies. So my chi-square
487
00:42:18,800 --> 00:42:22,400
statistic is
488
00:42:22,400 --> 00:42:32,260
41 minus 51 squared divided by 51 plus 61 minus 51
489
00:42:32,260 --> 00:42:44,160
squared. 561 plus 109 minus 99 squared 99 plus 89
490
00:42:44,160 --> 00:42:47,040
minus 99 squared.
491
00:42:49,080 --> 00:42:53,140
That will give 5 point.
492
00:42:57,260 --> 00:43:01,760
So the answer is 5.9418.
493
00:43:03,410 --> 00:43:06,210
So simple calculation will give this result. Now,
494
00:43:06,450 --> 00:43:10,370
next one, referring to the same information we
495
00:43:10,370 --> 00:43:15,890
have at 5% level of significance, the critical
496
00:43:15,890 --> 00:43:18,510
value of that statistic. In this case, we are
497
00:43:18,510 --> 00:43:22,690
talking about 2 by 2 table, and alpha is 5. So
498
00:43:22,690 --> 00:43:28,130
your critical value is 3 point. So chi squared
499
00:43:28,130 --> 00:43:31,610
alpha, 5% and 1 degrees of freedom.
500
00:43:35,000 --> 00:43:39,220
This is the smallest value when alpha is 5%, so 3
501
00:43:39,220 --> 00:43:41,440
.8415.
502
00:43:46,160 --> 00:43:51,760
Again, degrees of freedom of this statistic are 1,
503
00:43:52,500 --> 00:43:53,800
2 by 2 is 1.
504
00:43:56,380 --> 00:44:01,620
Now at 5% level of significance, the conclusion is
505
00:44:01,620 --> 00:44:01,980
that
506
00:44:06,840 --> 00:44:16,380
In this case, we reject H0. And H0 says the two
507
00:44:16,380 --> 00:44:20,800
variables are independent. X and Y are
508
00:44:20,800 --> 00:44:25,860
independent. We reject that they are independent.
509
00:44:27,380 --> 00:44:33,200
That means they are dependent or related. So, A,
510
00:44:33,520 --> 00:44:36,680
brand awareness of female TV viewers and the
511
00:44:36,680 --> 00:44:41,380
gender of the spokesperson are independent. No,
512
00:44:41,580 --> 00:44:45,200
because we reject the null hypothesis. B, brand
513
00:44:45,200 --> 00:44:48,340
awareness of female TV viewers and the gender of
514
00:44:48,340 --> 00:44:53,140
spokesperson are not independent. Since we reject,
515
00:44:53,380 --> 00:44:58,330
then they are not. Because it's a complement. So,
516
00:44:58,430 --> 00:45:02,810
B is the correct answer. Now, C. A brand awareness
517
00:45:02,810 --> 00:45:05,450
of female TV viewers and the gender of the
518
00:45:05,450 --> 00:45:10,550
spokesperson are related. The same meaning. They
519
00:45:10,550 --> 00:45:15,470
are either, you say, not independent, related or
520
00:45:15,470 --> 00:45:15,950
dependent.
521
00:45:19,490 --> 00:45:24,930
Either is the same, so C is correct. D both B and
522
00:45:24,930 --> 00:45:28,970
C, so D is the correct answer. So again, if we
523
00:45:28,970 --> 00:45:31,650
reject the null hypothesis, it means the two
524
00:45:31,650 --> 00:45:36,990
variables either not independent or related or
525
00:45:36,990 --> 00:45:38,290
dependent.
526
00:45:40,550 --> 00:45:46,630
Any question? I will stop at this point. Next
527
00:45:46,630 --> 00:45:47,750
time, inshallah, we'll start.