It is never a good idea to categorize quantitative, numerical data.

The best (to my opinion!) solution is to find a sensible statistical model relating the variables. Such a model has a "deterministic part" describing the functional relationship and a "stochastic part" to model the unexplainable variation in the data. This requires a careful analysis of the data and a good knowledge about the topic (the underlying theories, i.e. expertise in your field of reasearch) and some experience with statistical modelling. This task is not always simple and it might turn out beneficial to cooperate with a local statistician.

The habit of "testing this and testing that" seems not very useful to me, because often the tested hypotheses are not really clear, no relevant alternative is (and can be) specified, and teh result of a test does not tell us anything about the effect sizes (which are more [most?] interesting) . Shaping the data to fit it into a test is usually the worst one can do. The "significance" you get may help to get your stuff published, but it usually won't help anyone to really learn something from this data.

Thanks for the answer Jochen. The problem i have is the sample size (n=50) with this sample size is difficult to make a more deeper sense of the topic. This study is a part of one big study.

I have more variables in the study, but the main researcher ask me to see the "relative relationships between Age, BMI and the foot posture index. The objetive is to find if there are relationship between the variables. and then the  *relative* increase/decrease of BMI and foot posture (ie underweight or overweight/obese).

I have categorized them because is more easy to classify the BMI as underweight, normal and overweight as the CDC classify BMI in children. Foot posture index, have a scale between the ranks i write in the last post. (the sacel goes to -12 to +12 and the ranks are -12 to -5 "highly supinated", -4 to 0 "Supinated", 0 to +5 "Neutral", 6 to 9 "Pronated" and 10 to +12 "Highly Pronated"... This is why i have categorized my variables. Categorizing them i can make a relationship with the categories... What do you think?

Thanks

Multinomial logistic regression if the DV is foot posture and ordinal logistic regression if DV is categorized BMI.

I have been with a statistician, he made me linear regresion, not more. But certainly i not need to find how independent variables affect the dependent variable (foot posture index), only relationship between the BMI and FPI ( this is the objective the main researcher ask me to do).

I know regressions give more information... But when can i use correlations test?

Another question is please, what will be the difference of use the two variables numerical or classify each measure of each children into rank and then compare ranks for categories? How can it affect to the study's results?

The question to answer is... Have the children with more BMI, a flatter foot?

Thanks, i'm little stuck... Two many ways to do.

Good responses

I would not categorize either variable if your purpose is to understand the relationship.  Linear regression is certainly a reasonable starting point.  However, quantile regression might also be useful - not only does it make fewer assumptions than linear regression, but it allows you to examine various levels of the dependent variable

However, learning to use linear regression appropriately takes more than anyone here can answer - entire books are written about it.

 Thanks Peter,

I will do some research, but i do not want to change the plans to the main researcher, i will try to use spearman correlation, but really, how is it he form to correctly use spearman's correlation Between 2 continous variables in the same group?. I need to do any more process than cdo two clicks in spss?

Thanks all for the help

I don't understand.

Sorry, Peter, i refer if you need to have any assumption before to do a Spearman's correlation. For example... a sample without normality distribution... or any type of variable treatment.

Another question: The Foot Posture test, is not a continous variable, i'm correct? it is ordinal because it goes from -12 to +12 and can not have decimal numbers, in each test included in foot posture index only can be -2,-1,0,+1,+2... I'm reviewing the data and type of variables to find any "novel" error. Then if it is ordinal... one numerical (ordinal) can be related with numerical (continous) to perform a Spearman's correlation test? Both are not normal.

Thanks

Foot posture is a nominal and bmi is an ordinal variable.

BMI is not ordinal, it is interval or even ratio.  Whether foot posture is nominal is unlear, but probably not.

Peter is right. But the categorize BMI from underweight to obese has an ordinal nature because it is categorized from low weight to high weight. Foot posture categorization:  "Supinated", "Netrual", "Pronated", "High Pronated" is a categorical variable. 

If you categorize BMI and foot posture, both become ordinal.  But categorizing continuous variables is nearly always a bad idea.

Ok Peter and Abdolghani,

then, can i categorize both variables and use for example crosstabs to describe the sample?

after that if we suppose we use the both variables without category :

BMI as interval or ratio, continuous and,

Foot posture, continuous (or discrete?) which only can be a number between -12 to +12 and sometimes have ,5 decimal because we did the mean between both foot.

And is assumed that they are not normal...

Is possible to see association (only positive, negative  or not association) with Spearman correlations? I'm correct?

( thanks to you, now I know I should do regression which have more power... But this time I have no option to change)

As an aside, the FPI is not a measure of flatness of the foot arch, rather it is a measure of pronation, which is not the same thing. If I had to use it (my preference is for the Arch Index rather than the FPI), I would categorise it into three groups (pronated, supinated an neutral) and use multinomial logistic regression. 

Dear Adrian,

i know the use of FPI, but I refer with flat foot the entire entity. Rear foot version, forefoot abduction, collapse of the arch ( incongruence of it)... Etc.as Redmond said in his paper. If foot is very pronated, probably is flat. The FPI measure the posture of the foot in static position or relaxed calcaneal position with this indicators. Arch index only measure the arch height by a foot print and classify them with that... Footprints can be ok, but for me ( in my opinion) some foots can lie ( foots with excess of pronation where lateral arch is not present in footprint and you see the footprint as pes cavus... But it is pronated,

thanks for the info, as I see, too many ways to do the same thing...

Dear Friends:

One last question

i have collected more sample and testing with shapiro wilk, both continous variables (FPI and BMI) and they are normal distributed now (p>0,05 in SW)

I do Pearson correlation then but i got p= 0,339 which i assume there is not relationship between  both variables (with this sample of n=25) intra-group, Scatterplot shows r=0,040.

It is ok or correct to make conclusions with this information? 

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

Now i need to compare this both variables between 2 independents groups. If the variables of the other group are normal too... which test will be suitable to compare both FPI between groups... t-test, one-way ANOVA? (both groups are independent: different ages and variables are the same FPI continous )

The problem i'm having is that i'm not very sure if i'm in the correct way to resolve the objectives with the Test.

Now following your answers, i want to do a  regression to see how the predictors affect the dependent variable of FPI, If i want use the gender (nominal male=1 female=2) with the other continous variables, Do i need to make anything before to do the linear regression or should i use multiple regression?

Thanks for the help, it is much helpful for me.

Do an independent t-test for comparing the given variables. Do Ancova, if you have covariates. Apply simple linear regression one by one and a multiple linear regression. Besides, you should make a dummy variable for the gender: male=1 and female=0.

I do hope the following link will be useful:

Associations of learning style with cultural values and demographics in nursing students in Iran and Malaysia: https://www.researchgate.net/publication/280826829_Associations_of_learning_style_with_cultural_values_and_demographics_in_nursing_students_in_Iran_and_Malaysia 

Article Associations of learning style with cultural values and demo...

Thanks a lot Abdolghani

Now i have the things more clear. I will do Pearson's for correlation, multiple regression and t-test (to see differences between groups).

Your article is very useful for me, thanks.

question : i have two variables. the first is room temperature and second is shivering. they have three categories. which one i choose to investigate the correlation of that. chisquare or spearman?

It's a bad design of the study having these variables measured in three categories. Both variables could be measured on a metric scale. But that's a design issue.

Regarding the correlation: since both variables in your case are ordinal, spearman would be the better choice. But since the variables have only three categories, there will likely be a lot of ties and tha't where Spearman does not work well. The better altenative is Kendall's tau, particularily tau-b, that is specifically adapted to handle ties.