# Which statistical test should I use

I am currently completing my results of my dissertation, sadly I have little or no clue about stats or SPSS and am in desperate need of help.

My dissertation experiment is on delayed onset muscle sorness,my participants performed a workout to induce DOMS and its effects were tested 24 hours, 48 and 72 hours later. the test was a vas (pain scaled) scale to which I tested 3 stretches, sitting to standing and sprinting.vertical leap was also measured 3 times to gain a mean average.

Three weeks later the test was ran again with the same participants with the intervention of an ice bath straight after the workout, the same testing was then performed after 24, 48 and 72 hours again.

Which stats test do I perfom!? A independent t test? a dependent t test? and anover. Im very very confused
###### Who is Participating?

ProfessorCommented:
Well you can still compare outcomes individually with repeated-measures MANOVA - the difference is that it corrects for experiment-wise error rates.

In a sense, if you just conduct individual tests, you end up biasing your analyses toward finding statistical significance (which is bad).  After you've done MANOVA, you can conduct post hoc and contrast tests (which correct for that bias) when comparing individual times to other times (or even sets of measures to other sets of measures).

I suggest starting by reading these websites:
http://www.utexas.edu/cc/docs/stat38.html
http://faculty.chass.ncsu.edu/garson/PA765/manova.htm
http://www.statsoft.com/textbook/anova-manova/#mbetween
0

ProfessorCommented:
I'm a bit curious as to how you are producing a dissertation without any statistical training.  There are many possible ways to analyze this data, depending on the questions you are trying to answer.  What questions are you trying to answer?  What are your hypotheses?

Also, can you list out all of your variables?  From your description, I'm thinking you have 24/48/72 hour measures for 3 stretches + vertical leap, over two times points (24 total variables)?

If you're interested in changes over time, you will likely need some sort variety of repeated-measures MANOVA.
0

Commented:
Sounds like a job for a proper Design of Experiments tool.
0

Author Commented:

Sadly I was untrained for statistics as I took an overseas exchange during my second year and I have had little help since.

My hypothesis is: It is hypothesised that if delayed onset muscle soreness is relieved by ice water immersion, then basketball players are able to recover quicker.

my variables are on the  - VAS scale 6 stretches, 2 spints of a basketball court,
- Swelling of the quadricep, inches, 2 measurements
- Verticle leap, three jumps to find the mean in inches
These variables were measure 7 times in total, once for the baseline, 3 times after the inducing doms without an ice bath, 3 times after inducing DOMS with an ice bath. the measurements were 24hours apart.

I guess I am trying to analyse the variables against teach other, ie 24 hour results with ice immersion without ice immersion. The experiment again was to see if there was a positive effect n the treatment of delayed onset muscle soreness with ice water immersion in basketball players

Does this help explain things better, thanks again for your help

0

ProfessorCommented:
That's a fairly complicated design, and there are a number of ways to go about it.  You need to first test though that all of your variables are normally distributed - compute skewness and kurtosis statistics (Descriptives, in SPSS), and also look at histograms to make sure each variable is roughly in the shape of a bell.

After that, it depends again - are you interested in outcomes at each time point individually, or specific tests?  For example, you could conduct a single paired-samples t-test to examine differences between the 24h with bath and 24 without bath conditions. But that's probably overly simplistic.

My initial guess would be to conduct a repeated measures MANOVA with a repeated-measures factor for each of your DV outcomes (VAS/swelling/leap) - Repeated Measures is under Univariate Model in SPSS, I believe.

By the way, there's an SPSS zone on EE.  You probably should cross-list this there.  :)
0

Author Commented:
aha, well now I need your opinion! I am specifically interested in the outcomes of each time point individually as delayed onset muscle soreness changes over time. So i can reference if the ice water immersion was effective at certain times. however if this is seen as too over simplistic I obviously want the best grade I can possibly acquire on my dissertation. However if this is what I'm looking to establish and other studies have used similar ideas would this still be over simplistic?
0

Commented:
If I understood your description properly, technically speaking you have these elements in your experiment:

1. Two variables:
- A qualitative variable, namely whether or not an ice bath has been applied
- A quantitative variable: Time passed since inducing DOMS.

- VAS scale 6 stretches, 2 spints of a basketball court,
- Swelling of the quadricep, inches, 2 measurements
- Verticle leap, three jumps to find the mean in inches

Correct?
0

Author Commented:
correct
0

Commented:
Some thoughts about measures. While many of the measurements appear to be ratio measurements they may in fact be at best ordinal. VAS is a highly subjective ranked scale.. ordinal at best.  Use tests for ordinal measures there. (Mann-Whitney is a good one) I would also suggest making the comparisons pair wise which would then lend itself to a sign test.  Less sensitivity but it eliminates a lot of bias.   One can rearrange the pairwise  data to do the same thing as a repeated measures test. (same with other response variables)

Vertical leap appears to be ratio, but the distribution may be non normal. I would suggest correcting for individual bias by computing the differences pair wise (same subject, different conditions) to get a delta. The delta stands a better chance of being normally distributed  and unbiased, even if the individual measures are not normally distributed.

Swelling of quadriceps... same as vertical leap.

Once you have calculated indexes or differences you will have variables which should behave in a manner that allows hypothesis testing.

Sometimes one line of analysis does not work out because of high experimental error (random error). An alternative method of analysis should always be considered. In a similar situation I took each measure as a rank. ( min to max). I then normalized the ranks so that each set of responses had ranks ranging from 1 to 10.  I would  then combined the ranks on an individual by individual basis and added all the ranks of the measures (VAS, Leap, Quadricep) [with and  without ice bath in your case] and turned these sums into a single index which I called a response index.  I then conducted a repeated measure analysis using these index values.   A word of caution. Never look for the set of statistical tests which give the "best" answer. Look for the set of tests which give the most sensitivity to small differences for the data you have and once the data fits the model. Take the results. Good, bad or indifferent. As the saying goes "If the shoe fits... wear it "

Final suggestion. Use R+ or MathStat or better yet SAS:JMP  to do the calculations. I have had very bad experiences with SPSS. To me, its main failing is that while it will allow a user to specify the measurement type (ratio/continuous, Interval, Ordinal, Nominal).. it will then ignore that information and allow the user to calculate results using incorrect tests and methods.   I won't try to do a sell here.   Just a consideration. Our organization uses SAS and SAS:JMP exclusively (with a bit of Excel for really simple stuff) Think about it.
0
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.