Statistically significantly different percentages, I do not know how to answer this.

#1
Adult Type 1 diabetics were randomly assigned to standard care or participation in a six week workshop lead by a diabetic specialist nurse. Of the 50 diabetics randomly assigned to the workshop, 4 had ED visits for hypoglycemia in the following six months. Eight of the 50 diabetics in standard care had ED visits for hypoglycemia in the following six months.

Nurse workshop .08 (95%CI .032 to .188)
Standard care .16 (95%CI .083 to .285)

Was the rate of ED visits for hypoglycemic events statistically significantly different at .05 level between the two groups? YES or NO
How do you know?

My thought answer is "NO" because: The way to determine if things are statistically significant is by calculating a p-value. The confidence interval is an identified range of possible values “p” might take. Because the point estimate falls within the confidence interval, it is an automatic accepted (not rejected) p-value. Since both of the point estimates fall within their CI, the two groups are not significantly different.

Is this correct? Am I thinking about this all wrong?
 

Karabiner

TS Contributor
#2
Re: Statistically significantly different percentages, I do not know how to answer th

You have to perform a dirct comparison, i.e. to crosstabulate the variables "group" (standard versus workshop) x "ED visit for hypoglycemia" (yes/no) and perform the appropriate test, maybe Chi² or Fisher's exact test.
Since both of the point estimates fall within their CI, the two groups are not significantly different.
Sorry, but this doesn't make sense. Every point estimate MUST fall within its own CI.

With kind regards

K.
 
#3
Re: Statistically significantly different percentages, I do not know how to answer th

I found a t-value based on the two points using the formula:
Z(alpha)/2xSE(p1-p2)
This gives me a range value of (-0.047-0.21). Can I now say because the point of no difference (0 of 1) is included in this range it is not significant?
 
#4
Re: Statistically significantly different percentages, I do not know how to answer th

You have to perform a dirct comparison, i.e. to crosstabulate the variables "group" (standard versus workshop) x "ED visit for hypoglycemia" (yes/no) and perform the appropriate test, maybe Chi² or Fisher's exact test.

Sorry, but this doesn't make sense. Every point estimate MUST fall within its own CI.

With kind regards

K.
Exactly. I'll add on a little for the OP. In general, you compare the hypothesized value from the null hypothesis with the CI if you are using a CI to essentially conduct a test of hypothesis. This may be what prompted the confusion.