Can't figure out what test to use.

#1
Hello,

I have tried all kinds of test in spss in order to find out if there are any significant difference in compliance between different citys.
I have 16 citys and for each city i already calculated compliance in percentage.

I want to see if there are significant change in compliance (which it seems to be) but when i use kruskall wallis in spss it turns out to be 0.451 no matter what kind of numbers i put, so something is wrong, but what?

name of citys that i translated into numbers eg london = 1, dublin = 2 etc

1. 0.56
2. 0,31
3. 0.90
so on until 16

Grouping variable is 1 to 16 and test variable is compliance.

what do i do wrong?

please help me
 
#2
Study was to see how many people attended to ultrasound test.

Let us say that we sent 100 invitations in london and only 50 persons showed up then the compliance was 50/100 = 0,5 (50%)

So i have already calculated the compliance for all 16 citys, and it differs alot so it seems very strange p=0.451 when I use Kruskall Wallis.

Do you understand it better now, otherwise please let me know

thanks
gunnar
 

hlsmith

Less is more. Stay pure. Stay poor.
#3
This is still a little fuzzy. Seems like this should be a categorical (showed up yes/no) by group (16) problem (chi square). You lose information when you use percentages before conducting the statistic.
 
#4
thank you for your answers, but i do not think its correct with chi square since, a chi-square goodness of fit test allows us to test whether the observed proportions for a categorical variable differ from hypothesized proportions. In my case I do not have any hypothesized proportions.

this is my data:
cities attended to screening did not attend
los angeles 123 52
london 586 100
new york 123 86
dublin 54 52
tokyo 120 58
stockholm 1235 852
copenhagen 127 25
helsinki 126 75
hong kong 789 542
paris 543 123
madrid 896 254
brasil 30 30
milano 25 23
roma 75 25
budapest 10 5

how do you do that with chi square?
 
#5
so the first numbers after the citys are people who attended and the numbers after the first digits after the space are people not attended.
 
#6
yes, still a little more information is needed.

So i have already calculated the compliance for all 16 citys, and it differs alot so it seems very strange p=0.451 when I use Kruskall Wallis.
Ok, I understand it. You are dealing with percentages. But what is your goal? To see if the cities' compliance rates were different?

If so,you can run a chi-square goodness-of-fit test to see if all the cities had significantly different compliance rates or not? But remember, in that case you should enter the raw data, not the percenatges.

----------------------

With that format of your data, it is easily a case of chi-square then.
 
#7
Many here prefer chi-square test. For me it is more natural to estimate proportions and use logistic regression (logit model). (That will be almost the same thing as the chi-square as the others are talking about.) When I ran your data with logit it was a clear significant difference in the sense that the hypothesis of all cities being the same could be rejected.

But I think the most simple and clear analysis for drgunnar (nice name) would be to calculate proportions and confidence intervals for them for each city.

So for Los Angeles:
P = 52/123

1.96*sqrt(p*(1-p)/123)

0.423 +/- 0.087 would be the estimate and its “margin of error” (“confidence interval”)