1a. the crude odds ratio equals (90 x 2235)/(245 x 493) = 1.67. The stratum specific odds ratios equal (20 x 1854)/(55 x 407) = 1.66 and (70 x 381)/(190 x 86) = 1.63. The crude odds ratio and the stratum specific odds ratios are almost equal. This suggests that coffee and tea consumption is probably not a confounder.
1b. The following two tables display the relationship between the potential confounder and both the predictor and the outcome.
               Cigarettes                         Pregnant
Coffee or Tea   per Day          Coffee or Tea   in Six Months
Cups per Day     0     > 19      Cups per Day     Yes   No
  0-2           2261     75           0-2        1909  427
  GT 2           467    260           GT 2        571  156

The expected values for these two tables are shown below followed by the components of the chi square tests of independence.
                      Expected Values
               Cigarettes                         Pregnant
Coffee or Tea   per Day          Coffee or Tea   in Six Months
Cups per Day     0     > 19      Cups per Day     Yes     No
  0-2           2080.5  647.5         0-2        1891.4  444.6
  GT 2           255.5   79.5         GT 2        588.6  138.4

                 Components of Chi Square Statistics
               Cigarettes                         Pregnant
Coffee or Tea   per Day          Coffee or Tea   in Six Months
Cups per Day     0     > 19      Cups per Day     Yes     No
  0-2           15.7   50.3           0-2         0.2     0.7 
  GT 2         127.5  409.7           GT 2        0.5     2.2

Chi sq = 603.173  p < .001       Chi sq = 3.64, p > .05

There is a very strong relationship between the amount of coffee or tea drunk and the amount of smoking. On the other hand, the relationship between coffee or tea consumption and ability to get pregnant is not significant. Since only one of these relationships is significant, coffee or tea consumption is not a confounder.

Return to the questions

2a. The complication rate for surgeries performed by residents on call was 122/1840 = .066, and the complication rate for those not on call is 237/4672 = .051. The expected values and components to the chi square statistic are shown below.
          Expected Values          Chi sq Components
               Complication                 Complication
Call Status      No    Yes     Call Status    No    Yes
    No        1738.6  101.4        No        0.2    4.2
    Yes       4414.4  257.6        Yes       0.1    1.6

The test statistic equals 6.15, df = 1, and p = .013. Reject the null hypothesis, and conclude that there is a relationship. Those who were sleep deprived have a higher complication rate than those who were not.
b. The stratum specific expected values and components to chi square are shown below.
                        Emergency
          Expected Values          Chi sq Components
               Complication                 Complication
Call Status      No    Yes     Call Status    No    Yes
    No         627.9   75.1        No        0.0    0.2
    Yes       1045.1  124.9        Yes       0.0    0.1

                   Not Emergency
          Expected Values          Chi sq Components
               Complication                 Complication
Call Status      No    Yes     Call Status    No    Yes
    No        1098.0   39.0        No        0.0    0.4
    Yes       3382.0  120.0        Yes       0.0    0.1

The test statistics equal .35 and .57. Fail to reject the null hypothesis. There is no relationship between call status and complication rate after controlling for emergency status.
c. The data relating emergency status to the predictor and the outcome are shown below.
                 Emergency                       Emergency
Call Status     Yes      No       Complication  Yes      No
  No            703     1137         No        1673    4480
  Yes          1170     3502         Yes        200     159
Chi sq = 111.64, df = 1, p<.001   Chi sq = 134.66, df = 1, p<.001

The confounder is related to the predictor and the outcome. This is an important type of confounder. If we fail to control for it, we conclude that there is a relationship between the confounder and outcome when there really is not.

Return to the questions



Return to the syllabus

Return to the home page
© J.Rice