NOMINALIZING DATA MEASUREMENT

The author suggests that with the onset of electrical computers more sophisticated statistical manipulation was also coterminous  with the push to make nearly all data

ratio level of measurement. Although early statitstics date back to the 17^th century, mid

20 th  century electronic computer and software pushed statitistic that may be comprehensive, but still fall short  of validity. Thus, one suggests that when one studies a subject, that triangulation occur and that one of the other  options  is a safe cautious nominl llevel of analysis. In other words a “a ratio analysis pushback.”

INTRODUCTION

One can read the history of statitstics, but for the American classroom and Academia in general, the mid 40’s through the early to mid 60’s one was introduce to host of new strategies of statistics. These could be done by other more primitive tools, but the effort was immense.

By the time of my graduate school training, a room was set aside, a computer larger than an SUV was located, and then glassed in. This grand machine analyzed data. We were introduced to a relative hard paper bubble answer sheet  follwing each number through 100.

Thus a 50 point midterm was quickly analyzed for the number of misses or errors and subtracted from the total of  100 %. Other vitals were listed at the bottom. The man in the computer room wore a white coat as we stood outside his territory. A secretary that had been at the university stood next to me. She was in tears. For all those semester of her work life, she corrected tests by hand. It was long and laborious. Again, she was watching “science” work it wonders.

As time moved on, the department got a computer that did a lot of tricks however if you made one strategic move, the  machine was ruined and needed to be returned to the factory.

Analysis became cheap and the machine more technologically durable. Prices came down

And the opportunity to do science became available to more and more faculty and students. Both master and doctoral dissertations were filled an analysis of many variables at once.

RATIO STRATEGIES

Levels of measurement  had different properties and strengths. However if you assume hard numbers, ratio analysis makes all the variables contious. More chances can be taken and the formulas can be elegant. Some tricks of the trade was to take nominal variables and recast them as dummy variables. Thus male is equal to zero and female is equal to one. Some might say that thus females are twice whatever males are. No, it means that they are made ratio.

Some interval variables again can be recast into ratio levels without much difference in outcome. Here we now have a multiple regression model with many variations from forward solution, stepwise, path, fixed effects and many others. Econometric models blossomed as well as sociological multiple analysis with beta weights that explained the amount of variance of each variable compared to other variables.

PUSH BACK

Not all in every department of the social methodologies (social science) was enthralled with these changes. There were some who thought that theory would suffer to others who were unfamiliar and uncomfortable with stats. The empiricists went from advocates to outright exhuberants of stats saving the world.

In the early 60’s, one meteorologist analyzing wind patterns noticed that if he stopped the machine and started it again without all the numbers involved and gently rounded upwards just a fractional point, the measurement of simulated wind went wildly in another direction. In other words a small butterfly like phenomna  wildly redirected the wind into another direction. Chaos theory was born. Further, econometric models so sophisticated in appearance coulod bring down or nerly bring down portions of the stock market. Cheating was also easier as others not familiar with the intracies and complexity of the formula and analysis were essentially confused.  Science failed again. The euphoria was over.

As time went on deviant cases analysis was helpful in looking ast small cases that did not fit the mainstream but could strongly effect it. Other examples suggested that such things as meta-analysis could be off as much as a third compared to the gold standard of  research rviews.

Thus triangulation came to be to protect the researcher and those who be impacted by the analysis. Behavioral economics was born. Regular case studies and participation observation (with stepping in and stepping out) came to be valued again. Triangulation promised that the sophisticated formujla driven multi-variate analysis would be followed by other quanitative and qualitative analysis.

AN EXAMPLE

Although my 2 year school now does extensive follow up on what happens to our alumni as well as attrition, way back in the early 90’s I did my own research on graduation rates.

The dependent variable was did the liberal arts students go on to graduate at a 4 year school. I used the proper research protocols and arranged my variables along the lines of the well known “Tinto model.” The sample was stratified removing those who were in 1 and 2 year degrees technical programs. They did not have the desire or funds  to move on to a bacelorate institution. The sample was random.

After using stepwise forward solution, I discovered roughly nothing. So I resensitized the data by collapsing all continous variables into more modest categorical variables. What I found is that dividing all of the numbers  by the median (to reduce skew.) I disvoered that those individuals that had a B average at my school, worked less than

10 hours a week, and liked their experience at my two year school went on to graduate

 about 5 years later. This accounted for roughly two thirds of the students.

NOMINALIZE DATA MEASUREMENT

The strategy is to take all variables that are in both the numerator and the denominator and divide them into categorical numbers of X and Y. A 2 by 2 table is created and the

Analysis is done by Chi-square. Multiple chi-squares can look at 3 or  4 results at once.

The multi-tabular method can form inverse pyramids so that a number of variables are controlled at the same time. Avoid Fisher corrected. It can give the appearance of  ratio analysis. Rather, recast and recast for various results placed in the tables. Or, one can

Analyze 4 variables at once with a 2 x 4 tables. However, this second met should be avoided, one can only look at many variables at once rather than comparing a variable when the others are controlled.

CONCLUSION

Other multi variate strategies may be used when more technical methods higher on the stat-math food chain are used. One can fear multicollinerity with thse forced ratio variables and thus triangulation can help that may uncover some useful information.

Multivariate analysis should not be dismissed but rather complimented by both qualitative and simpler quanitative strategies.

REFERENCES