nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Should I graduate from Big Class.jmp?

yusuke_ono

This is a sensitive topic. If you feel so, please tell me, e-mail to me (Yusuke.Ono@jmp.com), or tell indirectly to someone (if you want to tell me anonymously). I will remove this entry.

When I check some features in JMP, I almost always use Big Class.jmp at first.

But I personally come to think this dataset is not a good sample data for beginner.

(1) Human physical data may be inappropriate ethically. Modern statisticians like Galton, K. Pearson and R.A. Fisher gathered human’s physical and mental data for “race” improvement. The “scientific” movement is called “eugenics”. From the current points of view, this movement is unacceptable.

Some people even deny using Fisher’s Iris data. Please see the following blog and comments.

https://armchairecology.blog/iris-dataset/

https://www.meganstodel.com/posts/no-to-iris/

(2) Big Class.jmp is a fictional dataset, but some people might think that this is a real dataset. It may be a questionable research practice (QRP) not to say a dataset is a faked one even if it’s so.

Accepted answers

All comments

Phil_Kay

Hi @yusuke_ono ,

Thanks for talking about this.

I am also uncomfortable using this dataset. In fact, I usually make a joke about how creepy it is that we are tracking the height and weight of school students.

In any case, I think we should be able to find better datasets that are relevant to our science and engineering users.

I also agree that we should be honest when a dataset is a faked one. I make a point of telling people when using simulated data. They understand why we can't always use real data. We should always be honest with our customers.

Thanks again,

Phil

MikeD_Anderson

I haven't had anyone complain about it. It is a useful data set in that we can send it to customers and they can prototype questions back to us without issues. It's also simple, approachable, and incredibly flexible.

That said, if you're uncomfortable, then you don't have to use it. The only times that I use it these days is when I need something to just show the mechanics of something (which it is really good for). I know other SE's just change the column names and recode where necessary to use it in other contexts. Since it's synthetic, it's fine to tinker with the data set however you want.

If you have an alternative data set that you think we could use in place of Big Class, you should approach product management and they can help you get the process started to have it reviewed. I felt strongly enough about Fisher's Iris data that I did just that.

Fisher's Iris data has been losing popularity because of his views (which you mentioned). I've been using the palmer penguin data as a stand in for Iris because I was uncomfortable with the association. Along the way, I found out that Palmer Penguins is probably a superior data set in many respects. It would also be a good stand in for Big Class if you want an alternative. I've uploaded it to JMP Public: https://public.jmp.com/packages/xQKykD2bwdwNwh7GxqBpL

yusuke_ono

Thank you very much for your reply, @Phil_Kay-san

I don't know the history of Big Class.jmp well, but I guess it comes from "SASHELP.CLASS", a sample dataset in SAS System. SASHELP.CLASS has 19 rows. (Big Class.jmp has 40 rows.)

In the documentation of SAS sample datasets (SAS document), the following explanation is written.

The Sashelp.Class data set provides information about a small fictitious class of students. Variables include Sex, Age, Height, and Weight. This data set is frequently used in SAS documentation to illustrate basic SAS coding. ...

As far as I remember, SASHELP.CLASS has already existed in around SAS Version 6.03. When I was an instructor for SAS training, I always used this dataset in the statistics course.

The dataset has a long history, so it might not be easy to graduate from it (at least for me).

yusuke_ono

Thank you very much for your response, @MikeD_Anderson -san

I agree that this kind of physical measurement data sets are useful for explaining basic statistical methods for beginners. I guess it's because many basic "modern" statistical tools (like histogram, variance, coefficient of variation, IQR, correlation, regression,...and so on) have been developed and customized for analyzing such datasets. But I also think that usefulness is different from ethics.

Even when Big Class.jmp is useful for explaining statistics intuitively, it might be ethically inappropriate to use it.

As far as I know, it is relatively recent for many statisticians to look back at negative history in statistics. In 2021, the session for “R.A. Fisher and Eugenics” have been held in JSM. In 2020, COPSS dropped Fisher's name form their award. In 2021, UCL released inquiry into the history of eugenics at UCL. In 2020, Cambridge university removed stained glass window of Fisher's Latin square.

Statistics must be useful for eugenics, and physical measurement data must be useful for explaining "modern" statistics. But it might be against ethics.

This is just my personal highly biased opinion, and it's not popular.

Peter_Polito

Thank you for bringing this up @yusuke_ono! The BigClass dataset makes me uncomfortable, I don't use it. Although the data is artificial I can't help but consider how I'd feel if my own kids were a part of a study like this, or if a note came home from school asking if they could participate. All statistics aside, it feels invasive.

yusuke_ono

When I answered a survey by National Academies yesterday, I found the third reason why I personally think "Big Class.jmp" is out of date. The survey has the following three features.

(a) The demographic questions (like your age) in the survey are optional. They asked "Are you willing to answer additional demographic questions?". You can select "No".

(b) One question asks "gender identity or identities", not "sex".

(c) Questions which ask "pronouns" and "gender identity or identities" are multiple responses. It means that you can "select all that apply" for these questions.

I had not noticed the concept like (c) before I answered this question, but I now think it must be good idea to make these questions multiple responses. Some people may have only one identity, but the other may have several identities. Although the situation must be very different, in Japanese, there are several pronouns expressing "I" or "me" (e.g. "Watashi" and "Boku"). In the music album, Hitsuji Bungaku's "12 Hugs (like Butterflies)", they use "Watashi" in one song, and do "Boku" in the other song. I myself also sometimes use in "Watashi", and sometimes use "Boku".

Many statistical concepts like correlation, variance, regression,...etc came from eugenics by Galton, K.Pearson and R.A.Fisher. They measured many physical and mental features like height, cubit, and the volume of skull. And, they classified people, animals and plants. Although I cannot throw away all of their statistical techniques, concepts or applications, I think I can use (a little bit) more up-to-date ones, and I would llike to try to do so.

Mark_Zwald

I've never used Big Class in a demo for the following reasons:

It is too simplistic and does not resonate with our audience (scientists and engineers)
It doesn't tell an interesting story (or if it does it escapes me)

With that said, I'm always on the lookout for interesting and relevant datasets that tell a story and our policy of not using customer data makes this challenging as that is the data a customer will resonate with most. Looking into AI tools generating interesting/relevant datasets and utilizing the TRL, Mastering Webinars and Case Studies more.

alisa_h_lowery

I personally do not use it with customers. I use the SAT dataset in the sample data library to give a JMP overview & can introduce the same graphical & statistical concepts as Big Class. I have used a version of it with high school students in the context of "let's tell your own data story" and I have them complete a survey/MS Teams Form with demographic information like height, weight, gender, I download their own data & pull it into JMP & create some visualizations & statistical discoveries with it. I do word it something like "Please give me your height in inches. (Guessing is OK or making up something is OK too as long as it's realistic.)" & same for weight. I let JMP create an equation for weight based on height. And I ask for a volunteer who is willing to have the class measure their height & weight to see how well our equation did. There is always a difference which leads to a discussion about the "error" term that none of them have learned about or maybe we should have created equations based on some of the other factors we collected data on. I try to use it such a way in that it's something they can relate to, but not to make them embarrassed or uncomfortable about it. Mary Loveless & I have discussed using head & arm circumferences as an alternative, but I haven't tried it yet.