DECONSTRUCTING STATISTICAL ANALYSIS

ABSTRACT:  Using a very complex statistical analysis and research method for the sake of enhancing the prestige of an article or making a new product or service legitimate needs to be monitored and questioned for accuracy. 1) The more complicated the statistical analysis, and research the fewer the number of learned readers can understand it. This adds a mystique to the article.  A journal with an elaborate analysis of a problem with an assumption of hard numbers can hide the presentation in which apples and oranges are brought together. 2) The chance that the article will probably be published is increased for that which has the most elaborate analysis. 3) Science is not served by the strategy of embellishing analysis. The soft sciences should drop their name from science to research methodologies. By stepping down in the stretch for power and prestige, the research methodologies should receive more respect. Elaborate jargon filled stats and methods replaced by transparent and simple mathematics should gain the acclaim of the hard sciences rather than the reverse if social sciences answers are valid and reliable.

 1. INTRODUCTION:

a) The, most prestigious analysis generally assumes hard number theory and multiple regression. This procedure fogs the findings. When 2 of the 4 independent variables are ordinal, one is ratio, and one is nominal, all are summated together as beta weights. Nominal, ordinal, interval, and ratio are separate from each other and when combining the various number powers, it has to appear that all are ratio. That means the findings have a zero base line, can be added, subtracted, divided, and multiplied. Therefore,

Each integer is completely exclusive and equidistant from each other and related.  That means apples and oranges can be divided by pears. It could be junk science. Further the researchers generally have medical or related terminal degrees are doing the research. The mystique is added given the credentials of the researchers and the size and power of the corporation or university indicate prestige and authenticity. The final copy is embellished by a professional writer who intertwines science and exotic connotations that suggests complexity. (Snell, J. & Marsh, M 2012)

b) The other is chi-square. (math.huis.edu/java math/Ryan/ChiSquare.html)  Chi-Square is easy to learn and understand. Thus the glow of science is lessened. Chi-Square is thought to be much less robust and analytical than say Step-Wise Multiple Regression. However, what appears counter intuitive is that Chi Square is less robust and more blunt and yet can cover more numerical territory. The formula is rather simplistic looking and yet it can handle all number powers from nominal to ratio.  The outcomes of the two (step-wise and chi-square) are not exactly the same.  Stepwise   may show differences when given a measure on a dependent variable which has ordinal power. However, that does not necessarily make it superior; rather a number of variables including the structure of the formula may lessen or increase differences. Further, using step-wise, we take all kinds of risks, blur number properties and may create a false outcomes. (www.socialresearchmenthods.net/kb/dummyvar.php)

2. THE CHANCE THE ARTICLE(S) WILL BE PUBLISHED:

a) As a thought experiment, two papers on related subjects (one using step-wise and the other chi-square) are released to the same editorial board. All editors have backgrounds in the social or soft science. Without hard evidence, this author can surmise that many go into the field to teach students their area. To them, statistics is to be endured and understood. However, both the editors and the review boards know their stats, but do not cherish the field. They may not catch the scam.

b) The author is comparing one of the most sophist iced statistical analysis and one of the most elemental ones. When the two papers are sent which manuscript is more likely to be accepted? Is it the one with the nominal chi-square, or, the step-wise analysis? Which lends more prestige and the ability to attract even more analytical essays from top schools? This author’s money is on the step- wise. It looks great. It glows when Greek, English acronyms and numbers flourish across the page if the formula is included in the article.

The valid outcome could have been with chi-square, because step-wise has so many risks that can be covered and enhanced by the nature of the complexity of reading the statistical strategy. Chi-square is dated and simple. It was replaced with step-wise when the computer could be used to make thousands of calculations in a short time. This is the onset of Big Data. Scholars who may have been intimidated are likely to approve the article.(Silver, N. 2012 9-12)

c) Are there any errors with chi-square? Here the author believes that there is a metaphorical wind at his back.  Nathan Silver (2012) leans on Gauss and Taleb, N. (2007) assuming Mandelbrot both come to similar, but not same exact conclusions. The research methodologies used now are flawed. Nearly all the researchers did what they thought was right, but did not have right outcomes. Their mentors experienced the excitement of Big Data and were seduced by merging various number powers together and as Ioannidis discovered (with the help of the Bayer Corporation) reanalyzing numerous published articles in prestigious medical journals is that two out of three journal articles were wrong. Top articles in top journals were wrong. (O’Connell, J. 2012) Years of work and thousands of articles were wrong.  No one or few cheated. They followed the rules to get something wrong. Worse yet, some comparisons are made between the new and fashionable with other expensive strategies. In the mean time, the approaches that are effective but not profitable to sell are overlooked. Ioannidis and colleagues looked at thousands of papers a year.  The saddest is the “file drawer effect.” The researchers did not find differences. It tempts researchers to cheat to “discover” differences further” not significant” tell a lot. No relationship is a finding and should not be recycled but sent to a journal that publishes abstracts where differences are not found.   All of this, points to any area where non-ratio or mixed ratio data methodologies is conducted. (O’Connell, J. 2012)

Silver was 49 out of 50 times correct when predicting who would vote for Obama in the 2012 elections. When economist and rating agencies were predicting the future, almost no one called for a massive economic crash in 2008. How many econometric models does one have to do before economists lose the trust of others? Dr. Li as well as Wall Street fraudsters help brings down the market in 2008. Merton using his Portfolio Theory gained the man a Nobel Prize but the theory facilitated a downturn in the 1990’s. (Snell. & M. Marsh, 2012)

How long can we continue to make nominal dummy variables into ratio numbers( see at the end of references) Multiple regression assumes that each variable remains isolated and mutually exclusive? Can we continue to accept that? Can we  continue to assume that variables remain relatively calm in a chaotically ordered world? Additionally, what happens when a more elaborate model inadvertently sullies a simpler and more accurate model? Last, an outcome is argued among the elite because the model’s structure gave birth to a truly cloudy and complex outcome? The few at the top argue and the remaining folks in the field wait for a decision, because the answer is generally beyond nearly everyone and if consensus emerges who translate this to other top scholars in the field?

We now reach the point where ambiquity, complexity, lack of understandable relationships, mystical  numbering help us to escape into unwarranted swamp, where academic elite do not understand it and thus fear rejecting it. Do MBA programs still have Merton’s portfolio even when it has been demonstrated to be wrong?

Taleb,N. (2007), the other critic made money on the crash using Mandelbrot. Both Silver and Taleb did not believe their teachers. The author when taking an advanced, statistical and research analysis still recalls that micro- second when the professor held back a smile (one that was fearful)  when he said that nominal through ordinal numbers can be blurred together.

Additionally, we were in education and social sciences in the boom of Big Data (1973) and were just smart enough to learn the information, but not criticize it.

3) SOFT SCIENCES ARE NOT SERVING US AND NEED TO BE RE-LABLED RESEARCH METHODOLOGIES

Chapter one of many research methodologies make claims that may generally not be supported in the real world. Can a chapter in the book indicate how easy it is to be wrong? Can the author(s) indicate that the soft sciences or research methodologies struggle at times with statistical analysis because the hard science generally studies subjects that are ratio? That the previous sentence can be stated probably more simply.

On the other hand, what we are doing now should not be cast asunder.  There are times when all the data are ratio. If that be the case we have the strategy. It is step-wise.

 Last, simple qualitative analysis can be priceless. The qualitative may be able to stand on its own or mixed with both qualitative and quantitative. Disciplines can change.  We now see that nature  and nurture are coming together. At one time that did not seem possible. So, it may apply to statistical analysis. We still need multivariate analysis. However, simpler measures at times may be more accurate. This is the most promising leap toward validity. 

REFERENCES CITED

Math.hus.edu/(javamath/ryan/ Chi Square.html

O’Connell, J.(2012) Something doesn’t add up,  Stanford Magazine, May-June, 1-5

Silver, N. (2012) The signal and the noise, why so many predications fail- but some don’t, New York: Penguin Press

Snell, J. & M. Marsh (2012) Multiple regression and its discontents, Education, vol.132, #3,  517-522

Taleb, N. (2007) the Black Swan, New York: Random House

(www.socialresearchmenthods.net/kb/dummyvar.php)

Dummy variables are structured as interval thus it can be used in multiple regression. This author  strongly disagrees. Please see an opposing this author Trochim, W.(2008) Dummy variables, Web center, Social Research Methods

.