It's suboptimal to have age in "age groups". If you know the years (or even the birth dates) then use them. Never ever discretize (put a numeric variable into groups).

Pearson assumes a linear relationship between age and response. This is likely not a good assumption.

Spearman assumes a monotone relationship. Better, but it may still be not good.

If you "find out if there is a relationship" you should not restict yourself to linear or monoton relatioships. Plot the data (response vs. years) and look at that scatterplot. If you see a pattern, then you actually found a relationship. You can use scatterplot smoothers or spline regression models to fit and show a smooth curve showing the (approximate) relationship.

Tests make not much sense here, since your analysis is exporative. If your data shows some kind of specific pattern, you may think of a model in which you could test a specific hypothesis (e.g., the response quickly peaks after 2 years and raises more slowly again after 6 years or so - you may then test if the data are sufficient to interpret the peak and/or the slow raising [notably on new data!]). The tests would need some idea about a reasonable distribution model for the response. I don't know how self-esteem is measured, but I think it will eventually be on some bounded scale (minimum to maximum possible value), in which case some beta-distribution or quasi-binomial distribution would be appropriate (after convenient rescaling).

PS: I wonder how you measure self-esteem. It's difficult in adults, but how can this be achieved in children? You mention a questionnaire. The children cannot answer a questinnaire. How do you find out that the answered of the parents (or who ever you are asking instead) is reliably providing information about the self-esteem of the children and not about their impression of the children's self-esteem, what is something quite different?

It's suboptimal to have age in "age groups". If you know the years (or even the birth dates) then use them. Never ever discretize (put a numeric variable into groups).

Pearson assumes a linear relationship between age and response. This is likely not a good assumption.

Spearman assumes a monotone relationship. Better, but it may still be not good.

If you "find out if there is a relationship" you should not restict yourself to linear or monoton relatioships. Plot the data (response vs. years) and look at that scatterplot. If you see a pattern, then you actually found a relationship. You can use scatterplot smoothers or spline regression models to fit and show a smooth curve showing the (approximate) relationship.

Tests make not much sense here, since your analysis is exporative. If your data shows some kind of specific pattern, you may think of a model in which you could test a specific hypothesis (e.g., the response quickly peaks after 2 years and raises more slowly again after 6 years or so - you may then test if the data are sufficient to interpret the peak and/or the slow raising [notably on new data!]). The tests would need some idea about a reasonable distribution model for the response. I don't know how self-esteem is measured, but I think it will eventually be on some bounded scale (minimum to maximum possible value), in which case some beta-distribution or quasi-binomial distribution would be appropriate (after convenient rescaling).

PS: I wonder how you measure self-esteem. It's difficult in adults, but how can this be achieved in children? You mention a questionnaire. The children cannot answer a questinnaire. How do you find out that the answered of the parents (or who ever you are asking instead) is reliably providing information about the self-esteem of the children and not about their impression of the children's self-esteem, what is something quite different?

How do you plan to measure self-esteem in children as young as 1? The statistical technique you can use will depend on the scale you can have for measurement of self-esteem- nominal, ordinal or interval. If you have ordinal scale of measurement that is a few categories of self-esteem then you can use logistic (2)or multinomial logistic (for more than 2). If you can have interval scale of measurement then you can use Pearson’s product moment correlation coefficient.

Harinder

Could you discuss more about dependent variable, i can help you

Harinder Mahajan , Jochen Wilhelm

Thank you for your response! Apologies on my end as I did not clarify in my question. The age groups basically represent the number of years my participants have spent in the sport. Not the age. I figured the way i structured my question was probably why there’s a bit of confusion.

It is years of experience in indoor rock climbing. Those were the “age groups“ I’ve mentioned.

The self esteem is sport state self esteem based on a 10-item scale. It’s a 5-point likert scale.

Basically what I want to observe is to see if there is a relationship.

As I spend more years in the sport, do I develop more self esteem of my own self-regard over the years?

Do novice and newbies in the sport have high self esteem? Or do the older folk who’s been doing the sport for years have more self esteem?

Rabia Almamlook You can see my response above. Apologies I did not cover all that is required.

I think they use poisson regression when there are several age categories as referred to because the spearman correlation will only determine in a general way and will not specify which age group.

Cristian Ramos-Vera What if i’m just trying to find out if there is a trend? That as we have spend more time, we have more/less self esteem?

Start with scatterplots of your variables. These should give you a good start. See for example: https://b-ok.cc/book/5825760/ed313f

Then google which regression should I use.

Best, D. Booth

As your data were divided into groups like age group, Kruskal Wallis is the better tool to find the correlation. This tool used Chi square in the analysis. Your results will be as numbers and percentage in the table in addition to the significance or p-value.