If anyone can help please!! In DESPERATE need of clarification!

#1
Hi Everyone

This is my first time on the site. Basically I am conducting some research for a dissertation I have to do as part of my course.

I will be investigating whether a variety of drinks thickened using powder thickeners changes over the period of one hour. I have already conducted the experiments. I used 3 different brands of thickeners. For each brand I thickened water, orange juice, orange squash and milk.

With thickened drinks (used for people who have swallowing difficulties) there are descriptors that the drink should meet so for each drink I tested whether they could

• Coat and slowly sinks through the prongs of a fork
• Leave a thick coat on the back of a spoon
• Cannot be drunk through a straw
• Can be tipped from a cup (to represent being drunk from a cup)

I conducted each fluid and thickener combination 15 times and have an excel spreadsheet of yes/no answers as to whether they meet the descriptors. I did this over the period of one hour testing the descriptors every 15 minutes.

When I went to see my statistics advisor, she recommended doing a binomial test find the percentage yes/no for each drink and thickener combination. She then suggested doing a chi sqaure test to see the difference between the zero minute time value and the 60 minute time value to see whether the consistency changed over an hour and whether it still fit the descriptors.

I am absolutely hopeless when it comes to statistics and is my biggest weakness. I have been researching chi squared tests and it doesn't seem to suggest that what has been said above is accurate. However, I have done the binomial test and it does show the percentages of yes/no's.

Any help or advice would be greatly appreciated.

Thanks

Lily
 

rogojel

TS Contributor
#2
hi,
you could look at two-way contigency tables (a fancy name for chi-squared test you need:) ).
Basically, you construct a table for each liquid and each property you test, with the rows passed, failed and the columns the times you tested the property and run the chi-squared test. It will tell you whether the time has a significant effect on the property or not.

In theory it is a bit problematic to have so many independent test because this increases the probability of a false positive. If you want to do this in more "correct" way you could have a single test per property with a three-way contigency tably adding drinks as a third dimension to your table, as described here:

http://web.ntpu.edu.tw/~cflin/Teach/Cate/06CateUEN05ThreeWayPPT.pdf
regards and good luck
 

Karabiner

TS Contributor
#3
I will be investigating whether a variety of drinks thickened using powder thickeners changes over the period of one hour.
So you are not interested in whether one brand is better than others?
Only whether they change over time?
When I went to see my statistics advisor, she recommended doing a binomial test find the percentage yes/no for each drink and thickener combination.
This doesn't address your abovementioned research question, but might
be interesting anyway. As rogojel already mentioned, the multiple testing
issue maybe is a bit of a problem.

If you wanted to put all brands & fluids into one analysis, you could perform
logistic regression analyses, but seemingly this is beyond your present
statistical skills.


She then suggested doing a chi sqaure test to see the difference between the zero minute time value and the 60 minute time value to see whether the consistency changed over an hour and whether it still fit the descriptors.
If you tested the very same drinks after 0 and 60 minutes, then this is a repeated
measures design and Chi² cannot be used. McNemar's test seems appropriate instead.

With kind regards

K.