29th July 2014 at 2:33 am #940JMember
This conversation is clearly beyond my level of knowledge…I am just a master’s student trying to properly analyze his data.
However, in a way I am happy with the discussion it has spawned, as its clear that you both truly love statistics, and I believe for the right reasons (although I get a sense that you Oscar in particular enjoy the debate aspect just as much…no?).
Anyhow, if the result brings us closer to the truth, and something good comes from it…particularly with regards to how social science approaches its questions, then I am all for it.
You see, my n of 18 is actually 18 very real young individuals diagnosed with schizophrenia (colloquially known as youth’s greatest disabler), and in my opinion, those suffering from this condition deserve the best that we involved in academia can offer. I think there are real and serious issues with how we go about analyzing and using research data in this population (and more broadly in psychiatry), and I am interested in ways to improve upon this. I think one way is to apply the the most accurate statistical methodology I can to the data I’ve gathered. If I compute findings based on false assumptions, and then go on to spread these findings (and publish them), then I would be doing a great disservice (well great might be a strong word…my project is far from great), but you get my point.
Anyways, thanks for the discussion. I myself am never one to shy away from a good debate…unless I have nothing of value to bring to the table…like in this case.
With that, I humbly bow out.
Jacob29th July 2014 at 3:41 am #939AnonymousInactive
well, for the case of your specific research question i would still side with Stephen because the alternative i present is much more complex (whether or not is more appropriate depends on who you ask). also, because you’re inquiring about a relatively straight-forward correlational design, i think everyone would agree on the same set of recommendations.
just as an intellectual exercise, even if we assumed by the graces of the gods that you could get your hands on a random sample, a simple power analysis with the usual power of .8 and alpha of .05 for a moderate correlation of 0.4 would still sit at N=29 or 30. and you only have 18. the fact of the matter is that even the strongest defender of null hypothesis testing would have to agree that results are unstable at such a small sample size, in which case a descriptive approach like Stephen recommended would be your best bet.
but in the thread where we discussed linear mixed effects models (multilevel models) where the design is complex, we do have very different approaches to the issue. i would really like for more people to jump into the conversation, but experience has taught me that people on methodspace tend to remain much more silent as opposed to other internet boards dedicated to data analysis/statistics like cross-validated or talkstats, where there tends to be a lot of active debate among data analysts/statisticians.
i guess i just have the problem that people rarely challenge me in my research group because of my technical background in mathematics/statistics. everybody assumes what i say must be correct because of my fancy math and programming skills. that, of course, becomes very big problem very quickly because it’s not easy to spot you own mistakes and implicit assumptions, even if you’re aware you’re making them. my real analysis professor used to call them “proof by assumption” where if you assume what you want to prove (like in a theorem) and then you shouldn’t be surprised that what you say is always true. that’s why i need people on the internet to point at holes in my reasoning so that i can learn as well.
well, both that and the fact that i spend too much time online but that’s a whole other story 😀29th August 2014 at 7:46 am #938Aizazullah KhanParticipant
Both of the correlation tools interpret in the same way, but you must sure about your data nature either it is ratio and interval scale data or qualitative data.
The pearson’s correlation is used when both the variables are quantitative while the spearman;s rank correlation is for qualitative data. Especially the data in likert scale29th August 2014 at 7:54 am #937Aizazullah KhanParticipant
hy Mr. Gorard would you like to give me your’s email. I am Aizazullah From Pakistan and i am the student of Ms. Statistics. I want to study your work. It will be my pleasure.
Thank You29th August 2014 at 8:59 am #936Stephen GorardParticipant
Happy to have a chat. I leave it to you( the researcher) to find my email. Hint – I am the only person with this name in the world.
- You must be logged in to reply to this topic.