Perception of non-parametric techniques as inferior?

CB

Super Moderator
#1
Hi guys,

I've recently worked on a project supervised by a lecturer in the school of psychology at my university. She's been looking over the article I've written and is concerned that my use of non-parametric statistical techniques through much of the data analysis is a problem, saying that publishers don't like to publish articles with non-parametric statistics - and that their use is perceived in psychology as indicating lacking methodology. This is the first time I've come across this notion, either within psychology or the broader social sciences.

I'm not particularly concerned about whether I should've used non-para techniques in the case at hand - it's a pretty open and shut case, a number of variables not being within coo-ee of normal distribution and so on. I was wondering, though, whether anyone else has come across this curious perception? I've always seen non-parametric methods as MORE methodologically vigorous, given that they require (by definition) fewer or no assumptions about the data at hand - assumptions that often aren't valid in real life data.

So - is this a peculiar misunderstanding on my supervisor's part, a commonly-held but inaccurate perception about non-parametric models, or can such models really be a tip-off of methodological weaknesses? :confused:
 
Last edited:

TheEcologist

Global Moderator
#2
Hi guys,

I've recently worked on criminology project supervised by a lecturer in the school of psychology at my university. She's been looking over the article I've written and is concerned that my use of non-parametric statistical techniques through much of the data analysis is a problem, saying that publishers don't like to publish articles with non-parametric statistics - and that their use is perceived in psychology as indicating lacking methodology. This is the first time I've come across this notion, either within psychology or the broader social sciences.
In one word nonsense. I know that some statisticians view non-parametric methods as superior as they leave less room for misuse and misunderstanding (of which I'm sad to say psychology and social science in general have a really bad reputation: that is with misuse of statistics: MISUSE OF STATISTICS IN SOCIAL-SCIENCES, Nature Volume 318, Issue 6046, 1985, Page 514 ).

I myself prefer distribution free stats.

Lacking methodology?

You cant change the nature of some datasets!

Size and age distributions in many populations are never normal (insect, trees). Parametric stats are of seldom use in low frequency high severity problems. If you sample parasitic or bacteria concentrations, you'll be very lucky to find normally distributed samples. Same goes for soil PH sampling. If you think about it there are millions of examples of count and bounded natural data where non-parametrics are of more use. Additionally the mean is not always of interest.

Here's a discussion on the t-test vs non-parametrics I had on an earlier date. This might help:

http://talkstats.com/showpost.php?p=17462&postcount=8

So - is this a peculiar misunderstanding on my supervisor's part, a commonly-held but inaccurate perception about non-parametric models, or can such models really be a tip-off of methodological weaknesses? :confused:
I'd say;
a misunderstanding on your supervisor's part that is fueled by a generally inaccurate perception about non-parametric methods.

You should try looking for info on 'robust statistics'. Non-parametrics often are robust statistics.

for example (copy paste from a unpublished report I have):

"Statistics as the median absolute deviation or interquartile range are robust measures of statistical dispersion, while the standard deviation and the range are not. With the latter two being highly elastic and strongly influenced by outliers."

"Medians are robust measures of central tendency (having high inelastically) in contrast to the mean; the median has a breakdown point of 50%, while the mean has a breakdown point of 0% (one sample can already influence it)"
 

CB

Super Moderator
#3
Thanks Marco, insightful comments as usual!

I know that some statisticians view non-parametric methods as superior as they leave less room for misuse and misunderstanding (of which I'm sad to say psychology and social science in general have a really bad reputation..
Sigh... I'd love to disagree, but you're exactly right about psychology and our other social science friends. I think the 'problem' in my own lovely field is that psychology tends to attract people who are sociable, empathetic, interested in emotions and helping people - but are not necessarily friends with numbers! I've seen intelligent, thoughtful psych postgrads terrified by statistical problems that business undergrads would find easy and science undergrads would roll their eyes at.

This isn't necessarily a bad thing - I'm a big fan of non-quantitative methods, and I think (in psychology at least) there's a lot more to be gained from a narrative description of a person or group's experiences, a discourse analysis, an ethnographic study - than yet another NHT study. But the dominant discourse in the field is that qualitative methods are inferior, so psychologists everywhere are involved in way of 'doing science' that they're probably not that suited to. I'd be far happier if far more psychologists were engaged in qualitative methods, leaving the odd strange fellow like myself who actually enjoys statistics to do quantitative stuff!

I'd say;
a misunderstanding on your supervisor's part that is fueled by a generally inaccurate perception about non-parametric methods.
Oh dear. I'd kinda hoped that she was unique in the misperception, but c'est la vie. I think I've done what I can to explain that non-parametric statistics are not parametric stats' spotty little brother - she seems unconvinced, but is willing to let me have my wanton way with the data.
 

TheEcologist

Global Moderator
#4
Thanks Marco, insightful comments as usual!
Sigh... I'd love to disagree, but you're exactly right about psychology and our other social science friends. I think the 'problem' in my own lovely field is that psychology tends to attract people who are sociable, empathetic, interested in emotions and helping people - but are not necessarily friends with numbers! I've seen intelligent, thoughtful psych postgrads terrified by statistical problems that business undergrads would find easy and science undergrads would roll their eyes at.
I think that is a good explanation on why things are as they are in those fields. Though I guess this would make a fine niche for you to exploit... :)

This isn't necessarily a bad thing - I'm a big fan of non-quantitative methods, and I think (in psychology at least) there's a lot more to be gained from a narrative description of a person or group's experiences, a discourse analysis, an ethnographic study - than yet another NHT study. But the dominant discourse in the field is that qualitative methods are inferior, so psychologists everywhere are involved in way of 'doing science' that they're probably not that suited to. I'd be far happier if far more psychologists were engaged in qualitative methods, leaving the odd strange fellow like myself who actually enjoys statistics to do quantitative stuff!
There's a time and place for everything, though I must disagree a little bit with you. Most sciences first go through a purely qualitative phase first before becoming more like the quantitative sciences we know today. Biology for instance 100 years ago (and before that) was mainly qualitative. Describing new species, collecting samples ect with the real big (quantitative) picture out of reach because the system is so complex. Darwins theory was qualitative, only later did we fully comprehend the quantitative basis for it. In my opinion biology has only just recently started to become a quantitative science. Now that people are starting to comprehend biological systems (genetics, evolution, ecology) we are realizing the full potential of quantitative methods.

Oh dear. I'd kinda hoped that she was unique in the misperception, but c'est la vie. I think I've done what I can to explain that non-parametric statistics are not parametric stats' spotty little brother - she seems unconvinced, but is willing to let me have my wanton way with the data.
:D

I hope it works out!