Jonathan -

The "new statistics," generally are not new at all, but some methods, like bootstrapping, and least absolute deviation regression, are more practical and/or can just be done better with advances in computers and software. But look at the foundations here. Whether OLS or absolute deviations are used, you still should not ignore the natural heteroscedasticity of the data. Weighted least squares regression can be far more useful. (I even saw recently where someone used OLS - which I generally do not like - and showed it worked better than I would have thought, for a particular type of application.)

One big problem in the area of the 'old' in statistics is the continued misuse of the p-value. I pointed that out decades ago, but it is still in misuse, though I think more people may be catching on to the problem now. (I have even seen on ResearchGate where it was noted that they are not even used as originally intended and we need to stop emphasizing them.) A p-value is a function of sample size, so at a given level, you can change your conclusions merely by changing your sample size. If you can use a confidence interval, you will generally obtain far more useful/practically interpretable results. If using a p-value, at least do a type II error analysis/power analysis/sensitivity analysis, for that situation.

The key is not old vs new, but to keep an open mind about everything. I like to say that experience does not count if you aren't 'paying attention.' Thus many with long 'experience' have not really benefitted by it and whether an idea is considered old or new, they don't process it. In order to process ideas, one has to start with a clear understanding of principles. Otherwise a 'new' innovation can just be "reinventing the wheel," often a square one.

In estimating totals, etc, for variables of interest for Official Statistics, I have seen establishment survey methods that grew more and more complicated over time, trying to obtain better results, when the real problem was that those who started the process did not understand basic principles in sampling and estimation themselves, and were mistaken in their original approach. Then all the tweaking done afterward to try to make things work better just made a bigger and bigger mess.

If someone claims an idea is 'new,' you may find it is just a rehash of something old, but it does not matter if it is old or truly new, or just a rejuvenated idea. It does not matter if it builds on an old idea or starts a new paradigm. What matters is that you consider any approach that might be fruitful and that you understand how it works to understand whether or not it actually applies well to the problem you are trying to solve.

No, you should not reject, or accept, a solution just because it is 'old' or just because it is 'new' (whether just purported to be new or actually new - which is rare). It does not matter. Just keep an open mind, and understand your statistical 'tools' rather than just rotely apply them.

Part of the problem you may have encountered is just that most people can be stubborn. Whether young of old, they fail to keep an open mind. Old folks like me can earn that reputation, but young people are not immune. - One day, coming home from work, I found a young woman standing in her yard across the street from mine. She was one of several students who were renting that house. She complained that she could not get into her house. Her key would not open the door. I thought she wanted help, so I came over and took her key to her door. Just as I started to unlock her door, she said "If I can't do it, you can't do it!" I quickly unlocked her door, handed her the key, and headed home, across the street. After she got over her shock, she called to me to ask how I did it. There was a trick to it that I told her, but the real answer was that I had learned from experience because I'd been paying attention. So ... just don't shut your mind off to varying possibilities, remember what you have learned, but still be willing to give everything a fresh look.

It is often very difficult for any of us to admit we might be wrong about something. It shakes our confidence. It is an humbling experience. But it is how we learn, and more importantly, I think, build character.

"I thought I made a mistake once. But I was wrong."     :-) 

(Note to the humorless: That was a joke.  Not original, but I like it.) 

Cheers - Jim   

Hi,

It's true but usually educational institutions take a while to catch up with the more modern science (including history, geography, economy, among others...). It usually takes an unofficial "probation" period for experimentalism and wide scientific community recognition of those new ideas for them to be included in the our "Universal Education" systems. It's not perfect but I understand the reason.

In advanced education however sometimes what you need to get is a professor which has broader horizons towards education. They appear but are rare. Since it's possible to create your own curriculum for a class in Universities these are the places where you'll be more likely to learn the "best so far" theories and methods.

Concerning statistics this will depend a lot on the purpose of the advanced course. Is it research (theory) oriented. Is it hand-on oriented? And also in what fields are you expecting to apply those methods? For instance I work in a research center that uses mathematics (including statistics) to study phenomena in geosciences. We're commonly considered researchers in spatial mathematics so we teach whatever we believe best for the students but there are problems. If I were to try doing scientific programming in a class, how many of the students would have enough background in programming to follow up? The same for advanced methods in statistics. Can you guarantee that their past educational experiences are solid enough to teach them something like that in a semester?

This is one of the reason why every course should have a reform every few years. In Europe bologna treaty helped a lot in this particular point (far less in others...).

Jonathan -

The "new statistics," generally are not new at all, but some methods, like bootstrapping, and least absolute deviation regression, are more practical and/or can just be done better with advances in computers and software. But look at the foundations here. Whether OLS or absolute deviations are used, you still should not ignore the natural heteroscedasticity of the data. Weighted least squares regression can be far more useful. (I even saw recently where someone used OLS - which I generally do not like - and showed it worked better than I would have thought, for a particular type of application.)

One big problem in the area of the 'old' in statistics is the continued misuse of the p-value. I pointed that out decades ago, but it is still in misuse, though I think more people may be catching on to the problem now. (I have even seen on ResearchGate where it was noted that they are not even used as originally intended and we need to stop emphasizing them.) A p-value is a function of sample size, so at a given level, you can change your conclusions merely by changing your sample size. If you can use a confidence interval, you will generally obtain far more useful/practically interpretable results. If using a p-value, at least do a type II error analysis/power analysis/sensitivity analysis, for that situation.

The key is not old vs new, but to keep an open mind about everything. I like to say that experience does not count if you aren't 'paying attention.' Thus many with long 'experience' have not really benefitted by it and whether an idea is considered old or new, they don't process it. In order to process ideas, one has to start with a clear understanding of principles. Otherwise a 'new' innovation can just be "reinventing the wheel," often a square one.

In estimating totals, etc, for variables of interest for Official Statistics, I have seen establishment survey methods that grew more and more complicated over time, trying to obtain better results, when the real problem was that those who started the process did not understand basic principles in sampling and estimation themselves, and were mistaken in their original approach. Then all the tweaking done afterward to try to make things work better just made a bigger and bigger mess.

If someone claims an idea is 'new,' you may find it is just a rehash of something old, but it does not matter if it is old or truly new, or just a rejuvenated idea. It does not matter if it builds on an old idea or starts a new paradigm. What matters is that you consider any approach that might be fruitful and that you understand how it works to understand whether or not it actually applies well to the problem you are trying to solve.

No, you should not reject, or accept, a solution just because it is 'old' or just because it is 'new' (whether just purported to be new or actually new - which is rare). It does not matter. Just keep an open mind, and understand your statistical 'tools' rather than just rotely apply them.

Part of the problem you may have encountered is just that most people can be stubborn. Whether young of old, they fail to keep an open mind. Old folks like me can earn that reputation, but young people are not immune. - One day, coming home from work, I found a young woman standing in her yard across the street from mine. She was one of several students who were renting that house. She complained that she could not get into her house. Her key would not open the door. I thought she wanted help, so I came over and took her key to her door. Just as I started to unlock her door, she said "If I can't do it, you can't do it!" I quickly unlocked her door, handed her the key, and headed home, across the street. After she got over her shock, she called to me to ask how I did it. There was a trick to it that I told her, but the real answer was that I had learned from experience because I'd been paying attention. So ... just don't shut your mind off to varying possibilities, remember what you have learned, but still be willing to give everything a fresh look.

It is often very difficult for any of us to admit we might be wrong about something. It shakes our confidence. It is an humbling experience. But it is how we learn, and more importantly, I think, build character.

"I thought I made a mistake once. But I was wrong."     :-) 

(Note to the humorless: That was a joke.  Not original, but I like it.) 

Cheers - Jim   

You need to drive the mini before the ferrari....

I agree with a lot of what was said here and I am sure we have all had articles rejected due to reviewers recommending antiquated usually classical analytic conventions.

This is probably one of the reasons I review way too much!

BUT if you have a look at R code and web based plug ins for GUI.....

The complex is becoming much more simple and accessible.

The times are a changing much quicker than previously. I am hopeful of that.

brad

"The key is not old vs new, but to keep an open mind about everything. I like to say that experience does not count if you aren't 'paying attention.' Thus many with long 'experience' have not really benefitted by it and whether an idea is considered old or new, they don't process it. In order to process ideas, one has to start with a clear understanding of principles. Otherwise a 'new' innovation can just be "reinventing the wheel," often a square one."

These statements could only come from a man of wisdom. You have my salute, James. Eddie

James is so right about the premise of learning... Learning from error and mistake and being BIG enough in our character to roll with failure and poor results especially when we age or build careers around topics...

remember some people become that invested that acknowledgement of errors is not an option. They have research centres and institutes... Patents and $sunk costs..

this is often an ageist thing as we build careers and profiles over time.... But James is also right...... It depends on the type of person you are. Aging with an open mind is not a bad motto.

publication and journals should also reward failure and negative results.... Often more important than confirmatory results.

and remember we are paid to give feedback and be experts and gain more status etc...become professors etc.. I do not like this aspect (except the $... I need some money for necessities and fun)..

keeping an open mind is a continuous battle..

brad

Your question starts with,"We all agree that statistic is the most important tool for scientists to proof (sic) if our hypothesis are true or false and therefore if our researches are good or bad."

Statistics does not prove anything. We use statistics to demonstrate aspects of our data. Proof is in the logical statements that develop the hypothesis and an assertion about the data that constitutes proof.

The problem is that by relying on statistics one can avoid the logic and assertions of proof. What does it mean to have a difference in means by the t-test? Is the difference large enough? What are the consequences of this difference? Will the difference in the yield of tomatoes be commercially significant or merely statistically different?

It is better to teach the scientific method instead of hypothesis testing without meaning. With hypothesis testing one can find a lame hypothesis and some statistical test that will show a difference.

Proper use of Bayesian methods follows the scientific method, but there are enough variations that look like hypothesis testing that science is easily sidestepped. 

Publications are most important to career development and most journals are happy to publish based on a p-value or confidence intervals.

Joseph has a good point that statistics are just tools.  The infamous p-value perhaps being the most misused. (Please see attached link.) 

Article Practical Interpretation of Hypothesis Tests - letter to the...