# Thread: significance test begin of puberty

1. ## significance test begin of puberty

Hello,

I am running a metaanalysis about the age of girls and boys at the beginning of puberty (Tanner-stages) and it would be great if someone could help me out with two questions about significance tests for mean/median value differences.

question 1:

In SPSS I have the data of the results of about 50 empirical studies. The two relevant variables are:
A) Birth year of the girls of the study either in the range "1975 to 1989" or in the range "1990 to 2003", so this variable is dichotomous.
There are 16 older studies ("1975 to 1989") and 34 newer studies ("1990 to 2003").
B) age at which 50% of the girls of the study reach puberty (Tanner-stage 2). These values (when girls reach puberty) are in the real world normally distributed.
Let's say the 16 older studies show a mean age of 10,7 years for the beginning of puberty (reaching Tanner-stage 2) and the 34 newer studies 9,9 years.
How can I find out if this mean value difference is significant, which test is appropriate?

question 2:
Let's say the old studies show a MEDIAN age of 10,7 years for the beginning of puberty (reaching Tanner-stage 2) and the new studies 9,9 years.
How can I find out if this MEDIAN value difference is significant, which test is appropriate?

It would be really GREAT if someone could help me. Thank you very much in advance!

Best wishes
Filip

2. ## Re: significance test begin of puberty

A) Birth year of the girls of the study either in the range "1975 to 1989" or in the range "1990 to 2003", so this variable is dichotomous.
There are 16 older studies ("1975 to 1989") and 34 newer studies ("1990 to 2003").
So a study from 1990 ist considered similar to a study from 2003, but dissimilar to a study from 1989?

Why don't you just use the actual study year instead of carrying out an artificial grouping? You could use correaltion/regression rather than group comparisons.

Let's say the 16 older studies show a mean age of 10,7 years for the beginning of puberty (reaching Tanner-stage 2) and the 34 newer studies 9,9 years.
How can I find out if this mean value difference is significant, which test is appropriate?
Why would you not take into account the sample sizes?

Anyway, you could do correlational analyses (Spearman correlation) between year and mean/median age.

With kind regards

Karabiner

3. ## Re: significance test begin of puberty

Hello,

I already have a correlation of Spearmans Rho and it´s significance in the article.

But I would also like to give the described mean and median value differences to show how much the values went down in that time period.

And I am not sure but I guess I have to add if this difference is significant so still my question would be which significance test would be appropriate (if there is one) for the mean and for the median difference.

"Let's say the 16 older studies show a mean age of 10,7 years for the beginning of puberty (reaching Tanner-stage 2) and the 34 newer studies 9,9 years.
How can I find out if this mean value difference is significant, which test is appropriate?

Why would you not take into account the sample sizes?"

I don´t know how to do this but I have in SPSS the information for each study how many girls were diagnosed.

Best wishes

Filip

4. ## Re: significance test begin of puberty

I already have a correlation of Spearmans Rho and it´s significance in the article.
oh, really? You did not mention that.
But I would also like to give the described mean and median value differences to show how much the values went down in that time period.
I do not find the idea of just 2 time period convincing at all. But regardless
of that, you can just display the according descriptive statistics for the 2
groups, if you like. There is no need to perform a test, thouh, if you have
already performed a correlation analysis. If you have a statistically significant
ordinal relationship between time and age, then testing age difference between
earlier and later times would be redundant.

With kind regards

K.

5. ## Re: significance test begin of puberty

hi,
fully agree with Karabiner, building groups like this is quite problematic unless there is a specific reason to pick a division. In a production environment this could be the introduction of a new machine for example, or new legislation, something that really makes the two groups potentially different.

If this is not the case, artificial grouping will make your point weaker.

regards

6. ## Re: significance test begin of puberty

"But I would also like to give the described mean and median value differences to show how much the values went down in that time period.

I do not find the idea of just 2 time period convincing at all."

The more time periods I create the less reliable are the results for these specific time periods because I just have about 50 studies and they diagnosed minors from different countries and different samples. So if I create smaller time periods the results of these few studies say more about the details of these specific studies than about the influence of the factor time. I also created three time periods and will publish these results but here it starts to get problematic.

Have a nice day

Filip

7. ## Re: significance test begin of puberty

Hello rogojel ,

"fully agree with Karabiner, building groups like this is quite problematic unless there is a specific reason to pick a division. In a production environment this could be the introduction of a new machine for example, or new legislation, something that really makes the two groups potentially different.

If this is not the case, artificial grouping will make your point weaker."

The grouping is kind of "artificial" and not natural. In Germany the age at menarche (first menstrual bleeding) went down from 17 years about 1850 to now below 13 years. So this reduction of the age when puberty starts and happens is a continuous development.

But as far as I know in social sciences it is common to create such "artificial" time periods. If there would be studies who show the age of menarche lets say in the year 1970 this would be good, but we don´t have studies like that because these studies have different results so it is necessary to combine studies. There are several metaanalyses about the age at menarche. These studies also create an "artificial" time period, they also combine studies from different years.
And it is not uncommon to compare the results of metaanalyses from different time periods. So from my point of view and according to my feeling about it what I do is not problematic. But I might be wrong.

Best wishes

Filip

8. ## Re: significance test begin of puberty

hi,
to give you an example of what I mean - is the difference between the groups influenced by the choice of your division year? - E.g. if you picked 1988 or 1990 or 1985 would the differences be more or less visible? It might make sense to look at this, no?

On a more general level, by grouping you destroy information - this can't be a good idea

Regards

9. ## Re: significance test begin of puberty

I am happy that I got this information too - does anoybody disagree?:

"for comparing means you should be good with using a simple 2 sample t-test to compare the means.

For comparing the medians you can try a Mann-Whitney U test to see if that gives you what you need - a good explanation is here: https://www.isixsigma.com/tools-temp...an-comparison/ "

10. ## Re: significance test begin of puberty

The U-test (a.k.a. Wilcoxon rank sum test) does NOT compare medians. If you want to compare medians, you can use the (sic!) median test.

The site you mentioned is a bit strange. That guy discusses normality, variance etc. with respect to the U-test, but the U-test is based on RANKS (ordinal data).

With kind regards

K.

 Tweet