Barclay 1

Statistics in the Social Sciences: Inferential Statistics as Rhetoric in Sociology

Brianne Barclay Wittenberg University 20 February 2004

I affirm that my work upholds the highest standards of honesty and academic integrity at Wittenberg, and that I have neither given nor received any unauthorized assistance.

___________________________________________________________

Barclay 2
TABLE OF CONTENTS: ABSTRACT INTRODUCTION PURPOSE DELIMITATION LIMITATION SIGNIFICANCE THEORY RESEARCH QUESTIONS RESEARCH HYPOTHESIS METHODS QUALITATIVE METHOD QUANTITATIVE METHOD QUALITATIVE ANALYSIS TERMINOLOGY EXAMPLE OF THE NULL HYPOTHESIS IN ACTION HISTORY WHAT IS RHETORIC IN STATISTICS IN SOCIOLOGY IN THE SOCIOLOGY OF STATISTICS IN REGARDS TO THE NULL HYPOTHESIS ACCEPTANCE V. FAILURE TO REJECT WHERE DO WE GO FROM HERE? QUANTITATIVE ANALYSIS: RESULTS ANALYSIS CAVEAT IN THE FINDINGS QUANTITATIVE CONCLUSIONS OVERALL CONCLUSION REFERENCES

Barclay 3 ABSTRACT: This mixed methods research looks at the usefulness of viewing statistics from a rhetorical perspective and what this means for the understanding of quantitative analysis in sociology. The study explores the rhetorical function of the null hypothesis in the practice and teaching of inferential statistics. Data is first collected from various literary sources associated with teaching statistics from a social science perspective as well as those related to the demystification of statistics and its theoretical assumptions. Literature, including Emile Durkheims work on rituals, Bertrand Russells theory of knowledge, and Max Webers concept of the ideal type, is then used to formulate and reflect upon the taken- for-granted character of the null. Themes from this data are then developed into an online survey (administered to a samp le of students and professors from a variety of fields, including Economics, Management, Mathematics, Psychology, Sociology, and Statistics) to test their ideas of the null hypothesis. In particular, inferential statistics is treated as a means to better understand interpretations and evaluations of significance levels and p-values. This paper is an exercise in the sociology of sociology and, more specifically, an exploration into the sociology of statistics, focusing on the rhetorical structure of inferential statistics.

Barclay 4 INTRODUCTION: If a picture is worth one thousand words, how many words is a number worth? How many words are necessary to convey the meaning of a single number? How many numbers are necessary to convey the meaning of a single word? These are all questions relevant to the study of statistics and to the importance of words and meaning in relation to this mathematically based field. Statistics is an important part of the social sciences. To say statistics is unnecessary or nonessential would be heresy. However, by lacking a sociological knowledge of statistics as a process of inquiry, we run a dangerous risk of misleading those with whom we seek to share our information. Through this thesis I look at the usefulness of statistics as rhetoric in sociology and what this means for the understanding of the field. This is an exercise in the sociology of sociology and the sociology of statistics. In 1992, I.D. John published an article in the Australian Psychologist titled Statistics as Rhetoric in Psychology. While psychology and sociology are different in their objectives, they share and blend basic principles as they are analyzed. The idea of statistics as rhetoric resonates in the field of sociology as much, if not more, than psychology. There is a unique language associated with statistics, and inasmuch as rhetoric is the art of speaking (or writing) well, there is a set of implicit guidelines one must follow when demonstrating statistical evidence to another. Another way of looking at the issue is as a counterpart of illiteracy. Being unable to read has been a dilemma for years in society, and when one cannot read the general public finds it a problem that must be altered and fixed. However, the problem of innumeracy (the inability to deal comfortably with the fundamental notions of number and chance (Paulos 1988:3) is just as present and just as serious. Unfortunately there is little of the enthusiasm for its change. John

Barclay 5 Allen Paulos has written a book by this very title (Innumeracy: Mathematical Illiteracy and Its Consequences) that discusses this distressing matter. People are quick to say they have difficulty determining sales tax just as though they were not able to fix an electronic gadget gone awry. Yet, mathematics makes up a substantial portion of the claims articulated by both professionals and non-professionals alike. How can it be that in a world where we rely on mathematics and science for our impressive technological advancements, we cannot balance a checkbook or cannot grasp the difference between one million or one billion? The importance of correctly identifying when and how to use statistical evidence is commonly overlooked. People have a tendency to think anyone who understands anything about numbers is apt to provide accurate results in their data (Turner 1987). It is commonly believed that anyone who tabulates numbers is a statistician. This is like believing that anyone who owns a scalpel is a surgeon (Hooke 1983:1). Instead statistical operations should be treated with care. However, not all data are, or should be, treated with the same amount of credence; they should be viewed skeptically before automatically assuming the corresponding evaluations are accurate. This idea of statistics as a form of rhetoric is virtually unexplored in sociology. Little research has been completed in other social science fields, but there is still some evidence that it exists as a topic of study. In sociology this is not the case, therefore this paper is important from a statistical, social scientific, and sociological perspective in that it takes each of these fields further than it has gone before. Based on statistical studies and evidence from other social science fields, one sees that often times gaps are present in statistical analysis. Research is conducted on the interpretations of statistical significance (Abelson 1995) and the proper use of terminology (Freedman, Pisani & Purves 1998), but there is no work done on the correlation of these two as they are tied to sociological rhetoric.

Barclay 6 On a collegiate level, in the departments of many social sciences, students are traditionally required to take a course in statistics. There is nothing out of the ordinary in this requirement, nor is it unsuitable. The problem lies in connecting the learned material to future research and studies. With only one course required for a social science undergraduate degree, all our knowledge about null hypotheses, chi-square tests, t-tests, and so on is to be taught in an approximately three hour a week course for around fourteen weeks in a semester. This may be a good idea, but it is not a realistic one. When this much material is covered in such a short period of time there are bound to be gaps in the teaching/learning process (Gal 2003). The idea of rhetoric is that the language is properly and effectively used in a specific context. In the departments of psychology and sociology, students are usually required to take one course in statistics. Statistical significance is conveyed through the knowledge and language provided for the audience. John says his study takes as a starting point for a considerationthe way in which inferential statistics serve a rhetorical purpose in providing epistemic authority in psychology (144). The goal of this study is to draw upon this idea and enter into the world of sociology using the same basic principle of inquiry.

PURPOSE: The purpose of this two-phase, sequential mixed methods study is to explore the involvement of the null hypothesis and the teaching of inferential statistics. In other words, I am beginning by using a qualitative method for exploratory purposes and following up with a quantitative method with a large sample of a specific population (Creswell 2003:16). The first phase will be a qualitative exploration of statistics as a form of rhetoric in the social sciences, in particular, sociology. Data will be collected from various literary sources associated with

Barclay 7 teaching statistics from a social science perspective, as well as those related to the demystification of statistics, more specifically the null hypothesis, as a constructive mode of communication. The emphasis of my work is on the qualitative study. However, the quantitative portion reinforces what is found through the theoretical resources. Themes from this qualitative data will then be developed into a quantitative survey to test the null hypothesis ambiguity and its function in interpretation. The intent is to show how an understanding of statistics as a form of rhetoric is critical to educational development as seen through the analysis of sample data from both professionals and non-professionals in the social science fields.

DELIMITATION: This study looks at the literature on the null hypothesis. It analyzes how what we know about the null is portrayed in statistics textbooks and journal literature, as well as other nonfictional writings. This study also considers how the null hypothesis, through a survey, is understood among the following groups among Wittenberg Universitys students and faculty: Economics majors and minors, Management majors and minors, Mathematics majors and minors, Psychology majors and minors, Sociology majors and minors, and Statistics minors who are all required to take statistics, and the professors in the corresponding fields.

LIMITATION: While this research is new and related to a virtually unstudied area, there is little information to guide the study. However, there is a vast amount of research on related topics and obviously not all the literature could be examined. Therefore only the books and articles available to me were included in the study that is a somewhat biased look at statistics.

Barclay 8 Also, the survey will focus on the students and faculty at Wittenberg University. Therefore there is no generalizability of the findings to other schools or other groups. Also since not all choose to respond, the findings are a non-random representation of those originally eligible. Therefore no inferences to the larger population can be made. In addition to the nonresponse of some asked to participate, those that did did not always complete every question. At times one or more questions was skipped, but since the survey cannot be re-administered to those individuals, the data analysis continued without everyones response to each question. In the same light, some respondents may have skewed interpretations to some questions, however there is no way to determine the frequency of this occurrence.

SIGNIFICANCE: This study is beneficial to the field of sociology because it aims at developing a sociology of statistics. The sociology of statistics is, in fact, a subset of the sociology of sociology. While neither area sends forth much literature into academia, the importance of both is critical to the understanding of the field as a whole. Inasmuch as the sociology of statistics ought to be studied, this new field can be understood as being created in three ways. It allows for a better understanding of statistics as a tool for rhetoric within the social sciences, in particular, sociology. It also helps us understand the principles surrounding the null hypothesis and the assumptions and interpretation based on the relative p-value. Finally it creates a better understanding of the process through which professors teach statistics to their students and how these students comprehend this newfound knowledge.

Barclay 9 THEORY: While the study of statistics as a form of rhetoric is not a common subject in the sociological theory, this does not mean that it is untouched by well-known theorists. Several such theorists have developed or elaborated on issues that can be directly applied to statistics. Bertrand Russell, in The Problems of Philosophy, says that when one thinks about the world of knowledge, it is customary to think of things in terms of what he does or does not understand. He does not think of how he knows it, just that he does. He further explains that there are two sorts of knowledge: knowledge of truths and knowledge of things. Here we are concerned with the knowledge of things and, in such a form, two further distinctions can be applied: knowledge by acquaintance and knowledge by description. Neither one is more important than the other for they simultaneously play a functional role in all aspects of society, but in the field of statistics it is one's goal to transform that by which he has description into that by which he has acquaintance. Knowledge by acquaintance can be interpreted as "anything of which we are directly aware, without the intermediary of any process of inference or any knowledge of truths" (1959:46). One could also say it occurs when "no further knowledge of it itself is even theoretically possible" (1959:47). Knowledge by description, on the other hand, occurs when "we know a description, and we know there is just one object to which this description applies, though the object is not directly known to us" (1959:47-48). Therefore in the realm of statistics it can be presumed that the null hypothesis and its interpretation correspond respectively with knowledge by acquaintance and knowledge by description. The null hypothesis is a term given to account for an analysis of inference that one understands completely. He knows it exists and sees how the calculations are carried out statistically. In mathematics he can completely understand calculations or manipulations of the

Barclay 10 data. The situation could be as simple as: Bill has three apples and gives one to Tom. How many does he have left? Or it could be as complicated as asking what percentage of chance one has of obtaining a sample of twenty students with an average amount of sleep in a given night being 6.346 hours if the estimated average is seven hours and the standard deviation of the hours is 1.397. Whether or not one fully understands the problem he is unwavering in the comprehension of the answer. In both cases the answer is two, and a person can know this in the fullest detail without lacking any further knowledge of it. However, when dealing with the null hypothesis, or more specifically, the p-value, there is no way to completely understand what the interpretation of the p-value is. Here is where knowledge by description enters the picture. In this category of knowledge one knows a description and there is only one object this description describes, even though we don't fully know the object (Russell 1959). One knows the definition of the null hypothesis, but can have no acquaintance with it because he or she cannot give a definitive explanation of the interpretation, as it does not exist. The theoretical concept behind its creation, however, must be fully understood in order to continue on through its study. However, one must accept certain p-values to be significant or else there would be no need to calculate (or no chance of reaching) an undeterminable result. For example anyone with statistical background knows that a p-value of 0.70 provides statistically significant evidence to refuse to reject the null hypothesis, but what about the p- value of 0.05? One would ask where the line between what is commonly known and what is reason for subjection lies. Do we know all we need to know about the null hypothesis? Do we know all there is to know about p-values if it is impossible to determine a single line between significance and non-significance? This is the point at which the literature on the subject offers no further clue toward another step. While Russell takes a stand on the evaluation of knowledge by description and

Barclay 11 knowledge by acquaintance in and of itself, the rest of the evaluation (in our case for the null hypothesis and p-value interpretations) is a theme up for more analysis. I think that regardless, though it might seem as though one merely knows it by description, in all actuality it is by acquaintance that it must be known (the above mentioned being the only characteristic of p- value interpretation). As Russell notes, "in addition to our acquaintance with particular existing things, we also have acquaintance with what we shall call universals, that is to say, general ideas, such as whiteness, diversity, brotherhood, and so on" (1959:51-52). By such a definition we must therefore include significance and p- value interpretation into our knowledge bank. At this juncture, once a theory is established for the usage of the null hypothesis and pvalue interpretations, another problem arises. As more and more people develop the common understanding of the numbers and their evaluations, these numbers will get them the recognition of their peers and predecessors. From this point it is an easy shift to rely on these numbers, fairly or non- fairly articulated, as the foremost portion of their work. At this point the processes become a habit and the meaning and value are pushed even farther into the background. Another way to describe this process would be to use Durkheims description of rituals. Durkheim (1965) explains the keys to a study of rituals. One point is that the language of a ritual depends on the shared meaning followed by a shared reality for a mutual understanding. In the practice of inferential statistics, rituals are performed as they consist of the shared understandings between the users and the readers. When explaining the null hypothesis, one should have a full grasp of the techniques used to obtain the results, yet this is often not the case. With the rise of computer technology and the ease in which results can be obtained, there is an overpowering urge to put to use any statistical evidence one finds, whether or not it is relevant to the problem at hand. An important feature in any inferential analysis of any subject is that the data and results

Barclay 12 are valid and pertinent to the study. When numbers are crunched with too much ease, there is the strong potential for the process to become a habit and incoherent, biased, or non-practical data to be found (McPherson 2001 and Creswell 2003). The kind of knowledge produced in this ritualistic thought must be adequately conveyed to the audience. Once again here comes rhetoric onto the scene. Dar explains that the statistical pedagogy overlooks the principles associated with philosophical perspectives related to the creation of knowledge and focuses on the recipes for predetermined forms of this knowledge construction (1987). By this process students neglect a complete understanding of their argument (once again the issue of knowledge by acquaintance versus knowledge by description) in favor of performing the ritual of the null hypothesis and inferential statistics. It is crucial for the researcher to contribute to the argument for a clearer and more precise explanation of his or her subject matter and the conclusions drawn. A final theorist from whose work we can draw upon for the context of this thesis is Jean Baudrillard and his description of a simulation. He recognizes that reality does not need to be "rational...it is nothing more than operational...no longer enveloped by an imaginary, it is no longer real at all. It is hyperreal..." (1999:482). Related to the null hypothesis, one could take this to mean the statistical realm does not need to be rational. It no longer is an issue of description or acquaintance, belief or knowledge, theoretical expression or a practical state. It is about a blending of the two qualities so there exists a knowledge so profound and so supreme it is hyperreal. The tangible result of a p-value is not necessary to grasp the greater meaning behind its interpretation. One only has to recognize the importance of this blend to fully acquaint himself with any further knowledge which can be derived.

Barclay 13 Such is the case in many everyday situations. We often do not know why something occurs, only that it does and therefore we must accept it. For example, we do not understand much of the fighting in the Balkans, but we do recognize it exists. Given the fact that we do not understand or relate to the tension does not make it any less real to other people who are in the midst of the crisis. While many may believe the null hypothesis is irrelevant to their lives, if the same beliefs are transposed to society, they would be quite clear. We must comprehend p-values in the same manner we comprehend most aspects of our lives. War is not rational, but is an issue with which to be dealt; we cannot be close- minded to say it does not affect us. Although we do not have knowledge by acquaintance with it, can we really say we have knowledge by description since much of what goes on is kept a secret during such times? At the same time we come to understand the knowledge, it is not useful to make any assertions because all that is needed is a knowledge that war exists in a hyperreal sense. The same is true with the null hypothesis, if the p-value is continually subject to interpretation, we can say it exists in the essence of hyperreality. Through the use of Russell, Durkheim, and Baudrillard the link between theory and the use of the null hypothesis and its corresponding p- value is evident. Deficiencies in this field exist, inasmuch as there are no theorists directly related to the sociological study of statistic s, but these four theorists leave an impression on this malleable subject.

RESEARCH QUESTIONS: In what ways is statistics a form of rhetoric in sociology? How has this been established in the social sciences? If the null hypothesis is the fundamental model of inferential statistics, what are valid interpretations of its corresponding p- value? How are (or are not) its assumptions

Barclay 14 accurately used and explained in the classroom? What is the overall purpose of studying the null hypothesis in general?

RESEARCH HYPOTHESIS: Null Hypothesis: All people, regardless of major, minor, or instructional field have the same understanding of interpretations and evaluations of significance levels and p-values. Alternative Hypothesis: Depending on ones major, minor, or instructional field, people have different understandings of interpretations and evaluations of significance levels and p-values.

METHODS: The design of this study is comprised of mixed methodology. As described in the introduction, mixed methodology allows the researcher to expand on the ideas or findings of one method with another. In this instance I use the qualitative method of textual analysis from the literature to provide a base for the quantitative method of survey design. By combining the two methods I can better analyze the data than if only one were to be chosen. The qualitative portion allows the reader to understand where the field of the sociology of statistics has gone in previous years and where it has the potential to go in the future. It allows a more comprehensive grasp of the theoretical principles surrounding the issues and relates to the heart of the problem, mainly, how and why rhetoric is important and how inferential statistics is an example of this. The quantitative portion allows the readers the ability to look at the how what is gathered in the first section is put to practical use. By only providing a textual analysis of this

Barclay 15 complicated idea, readers would miss how rhetoric relates directly to, not only to the selected sample (students and professors), but society at large. By only providing a quantitative portion, readers would miss why this is necessary for rhetorical understanding. Hence, in order to complete both sides of the rhetorical spectrum, a mixed methods approach is most useful. Another way of making this point is to visually see how this process works. This general illustration shows the compatibility of the two methods and how they interact with one another (see FIGURE 1).

FIGURE 1B ENEFITS OF M IXED M ETHODS APPROACH

Method: Qualitat ive Strategy: Textual Analysis Im portance: Histor y of Sociology of Statistics & Possibilit ies for the Future

Method: Quantitative Stra tegy: Online Su r vey Im p orta nce: Applicability to Sample & General Understanding of Soci etal I mporta nce

Q UALITATIVE M ETHOD : The approach of qualitative methodology chosen for this research is a textual analysis of the material related to the discussion of statistics and rhetoric. The first step in this process was to collect data useful for the topic. By conducting numerous online searches of Sociological

Barclay 16 Abstracts and EbscoHost, dozens of journal articles were located related to the teaching of statistics from a sociological vantage point. In addition to these articles, several books were found based on recommendations of professors in the field of statistics here at Wittenberg University. Through reading these data sources I was allowed a more comprehensive understanding of the field of sociological teaching of statistics. From this point I, once again with the help of other professors, located several statistics textbooks that set the stage for a grand overview of inferential statistics, in particular, the null hypothesis and its interpretations and evaluations. Finally several outside texts, typically referring to the issue of rhetoric, were located and read for analysis. While this portion reads as would a literature review, it is much more extensive and all encompassing than a traditional literature review for an undergraduate thesis. For this reason the literature mentioned in the introduction provided only a brief explanation of the material and will be developed in forthcoming portions of this paper. While the sociological articles, statistical textbooks, and other outside sources chosen provide an extensive list of authors, principles, and theories, obviously not every source could be examined. For this reason the ones I have chosen are a representative sample of the material available on the subject. As noted earlier in the limitations section, this field of the sociology of statistics is relatively new (the first reference by my observation was in 1973, The Sociology of Statistics: Measuring of Substandard Housing by George Sternlieb, but not regularly used thereafter). Therefore the selections have been chosen through my discretion as a reader and researcher, with the modest attempt to incorporate as much valid material as possible.

Barclay 17 Q UANTITATIVE M ETHOD: The approach for the quantitative methodology was to construct an online survey to be distributed to students and professors at Wittenberg University (a small, Midwestern liberal arts university). The survey was created at www.questionpro.com. This population consisted of 450 students and 36 professors from the schools most recent list of all majors, minors and professors (including full, associate, and assistant professors, as well as instructors) in the fields of Economics, Management, Mathematics, Psychology, Sociology, and Statistics. These individuals were chosen because of their involvement in majors or minors that require at least some knowledge of statistics. The goal was to survey both those with mathematical background and/or social science knowledge to determine if there was a difference, depending on their field of study or instruction, in interpretations and evaluations of statistical significance and p-values. The first step was to send a letter to everyone in the sample, explaining my intentions in conducting the survey and asking for their participation. The following table (TABLE 1) is a break down of those asked to participate. The total is slightly skewed due to two factors: some students are registered as both a major of a represented field, as well as a minor in another represented field (or a double major or double minor in two of the fields surveyed), and a few professors teach more than one subject. Of the 541 people listed in the total column, only 486 are unique individuals. Soon after, a similar letter was sent to them through e- mail. In this e- mail I explained the process of the online survey and provided them with a direct link to access the introduction and questions.

Barclay 18 TABLE 1POSSIBLE R ESPONSES BY D EPARTMENT Economics Management Mathematics Psychology Sociology Statistics Total Majors Minors Professors Total 32 10 6 48 184 40 8 232 39 9 6 54 108 15 7 130 35 26 8 69 N/A 7 3 10 398 107 38 541

NUMBER OF POSSIBLE R ESPONSES

The survey consisted of thirty-three questions, created by the research. The first question asked whether or not the individual had taken (or was currently taking) a statistics course. The rest of the opening questions related to the individuals background, asking about his or her major, minor, and/or area of instruction (this being the independent variable). The questions that followed dealt with the number of courses (and the field(s) of statistical learning) he or she had taken. The next set of questions asked for opinions of his or her basic level of understanding of statistics, in particular, the null hypothesis. Finally specific questions related to their comprehension and interpretation of the null hypothesis and its corresponding significant levels and p-values were asked. The last two sets of questions are related to the dependent variable. Although a formal pilot study was not conducted due to time constraints, the survey was administered to several students and professors at my convenience before being sent to the entire population of students and professors. The number of responses obtained in each field is broken down in TABLE 2. As noted above, the total number of unique responses of individuals is different than the number listed in

Barclay 19 the totals column. The total number of unique students who completed the survey was 81 and the total number of unique professors was 8.

TABLE 2OBTAINED R ESPONSES BY DEPARTMENT Economics Management Mathematics Psychology Sociology Statistics Total Majors Minors Professors Total 6 5 2 13 25 11 2 38 5 4 2 11 23 2 2 27 14 9 1 24 N/A 6 3 9 73 37 12 122

NUMBER OF OBTAINED R ESPONSES

Since the results were from a sample of volunteers, no randomization occurred in the planning. As such, the results from this analysis cannot be considered accurately representative of the whole population of those originally asked to participate. The analysis only shows the relationship among those who responded. The analysis also only shows the relationship of those who have taken a statistics course. The first question on the survey asks whether or not the individual has taken or is currently taking a statistics course. While a majority of the majors, minors, and professors in these fields have completed at least one of the required statistics courses, not all have done so. The survey is branched so that none of the individuals who have not taken at least one statistics course are asked any further questions related to the field. In total there were 107 people who took what they could of the survey. Of these 107, 89 students and professors had taken (or were currently taking) a statistics course. From these 89, the rest of the survey questions were administered and then analyzed. The total response rate of the survey was

Barclay 20 18.3%; the response rate of analyzable students was 18%; and the response rate of analyzable professors was 22.2%. Another point to notice is that no controlled experiment was conducted so the results do not attempt to explain any causal relationship in their findings. Any and all observed correlation can only be seen as that: a relationship. Lurking variables, such as the grade or achievement level of the individual, difficulty of the class, amount and/or quality of the material covered, number of years passed since completion of last related course, etc., were not accounted for in the survey. The survey did, however, account for confidentiality and anonymity. Through the options set forth by the website (www.questionpro.com), the responses are saved through their identification with the IP address of the computer on which the survey was taken. Because of the setup of computers on campus, it would be virtually impossible to identify a respondent through this IP address. Since more than one person could complete a survey from a computer, both confidentiality and anonymity are preserved. For a fuller understanding of the demographics of this study, the following pie charts (CHARTS 1, 2, AND 3) show the numbers, relative to one another.

Barclay 21 CHART 1D EPARTMENT OF STATISTIC COURSE CHART 2S TUDENTS YEAR

CHART 3N UMBER OF STATISTICS COURSES TAKEN (OR CURRENTLY TAKING)

Barclay 22 QUALITATIVE ANALYSIS: TERMINOLOGY: To begin an analysis of the sociology of statistics, one of the most important tasks is to understand the terminology behind statistics in the social sciences. Within the field there are two divergent branches of data analysis. The first is called descriptive statistics and the second, inferential statistics. Descriptive statistics are the values that describe the characteristics of a sample or population (Salkind 2000:367). It is a way of summarizing data into understandable quantities that can be compared to others in a similar manner. In the book The Little Prince, the main character describes the way grown- ups look at people. He says: Grown- ups love figures. When you tell them that you have made a new friend, they never ask you any questions about essential matters. They never say to you, What does his voice sound like? What games does he love best? Does he collect butterflies? Instead, they dema nd: How old is he? How many brothers has he? How much does he weigh? How much money does his father make? Only from these figures do they think they have learned anything about him. (de St. Exupery, as cited in Freedman, Pisani, and Purves 1998: 31) This quote provides insight into how people often view the world around them. In order to explain something to someone else, or just gain a better personal understanding, descriptive statistics are used to represent characteristics about a group. To reiterate Russells point, another way of describing the function of descriptive statistics is as providing a knowledge of acquaintance of the statistic. Inferential statistics, on the other hand, are tools that are used to infer the results based on a sample to a population (Salkind 2000: 368). In other words, one of the components of data analysis is to create a sample that is representative of the population. If this is the case, the conclusions made about the sample can be inferred to the entire population from which the sample was drawn. These statistics initially provide a knowledge by description, wherein the

Barclay 23 purpose of performing the calculations is to gather enough information about the sample (and hence the population) that eventually a knowledge of acquaintance can be obtained. In particular, the null hypothesis significance test has been a pivotal component of inferential statistics. The null hypothesis is a tool to analyze the difference in two or more groups. It looks at a descriptive statistic of a group (eg: the average of the variable to be analyzed) and compares it to the same descriptive statistic of another group. The purpose of the test is to determine whether or not there is a significant difference between the two samples. Since the samples are to be representative of the population, the same difference (or lack there of) between the samples can be inferred back to the populations. One example of this would be to look at the average score of 9th graders versus the average score of 12th graders on the ABC test for memorization skills. One would assume there would be no difference between the two samples of students from each grade. The only possibility for a difference would be due to chance if the null hypothesis were true (Salkind 2000). Researchersassume that nothing has been demonstrated unless the results can be clearly distinguished from chance or random error.in theory, all experiments are begun with a statement called the null hypothesis which states that there is no relationship between the independent variable and the dependent or response variable (Coolidge 2000:94). If the assumption is that there is no difference between the two variables, then it seems reasonable to say that any difference one can find would prove the null hypothesis to be false, insomuch as there is no relation between the two variables.

Barclay 24 EXAMPLE OF THE N ULL H YPOTHESIS IN ACTION : Imagine you want to know which tire on a car people think is more likely to go flat. Lets say you think the answer would more often be the front right tire. The population is the proportion of all people who select the right front tire as the one more likely to go flat; our sample can be the proportion of the guesses by the twenty people chosen from a list of professors at a local college. The null hypothesis is: H0 : ? = 0.25 (H0 being the symbol for the null hypothesis and ? being the symbol for the proportion of the population) since theoretically there should be an equal chance of all four being selected. The alternative hypothesis is Ha : ? > 0.25 (Ha being the symbol for the alternative hypothesis) since we think the actual proportion is higher than 25%. For our hypothetical sample of professors, lets say eight chose the front right tire as the one most often gone flat. The proportion of the sample is then 8/20 or 0.40. Through a few steps not necessary to explain here (but found in statistics textbooks), the test statistic (a value, in this case, computed to standardize the value of the sample proportion against that of the population proportion) is approximately 1.55. This value leads to a p-value (which determines the strength of the evidence against the null hypothesis) of 0.0606 (Rossman, Chance, & Lock 2001). (The idea of the p-value will be developed later in the thesis, but for now a note should be made regarding the interpretation that follows this sort of analysis: Professionals and non-professionals correctly articulate the terminology, yet often are stuck when it comes to correctly explaining its meaning in the larger scheme. Often the problem is not that individuals want to misinterpret the data as much as it is that they are unsure of how to use it correctly and in the confusion and frustration, results become distorted. Even some textbooks do not properly explain the usage or the involved steps. (In this thesis I do not imply that I have the best explanation--several books

Barclay 25 found in the bibliography can explain better, but I wish to give a brief overview for those who are unfamiliar with the process or need a refresher lesson.) The null hypothesis expresses the opinion that the only difference between a sample and a population is that of chance. It does not imply that there is no difference, only that what difference may be observed is quite minimal. On the other side is the alternative hypothesis. This says that the variation is not due to chance and that there is some outside force allowing for the difference (Freedman, Pisani & Purves 1998).)

HISTORY: Also necessary to understand the analysis of a sociology of statistics is a background knowledge of the null hypothesis in the social sciences. Franklin H. Giddings is known as the father of quantitative sociology in America. Although his studies used only the basic principles of what is used today, his knowledge of statistics usefulness has continued through the years (Raftery 2001). The null hypothesis began as a tool for agricultural studies after World War I and soon after integrated into the psychological realm (Rucci and Tweney, as cited in John, 1992). As time went on this sort of statistical testing, known as a significance test, became more popular in psychology and in Sterlings studies of 1955-56 he recorded the number of articles in several leading psychological journals (eg: Experimental Psychology, Comparative and Physiological Psychology, Clinical Psychology, and Social Psychology) and found there to be a substantial number of articles including significance tests in their findings and analysis. The percentages of articles that included significance tests as compared to the total number of articles ranged from 77% to 85%. Throughout the remaining years, statistics, in not only psychology but also other social sciences, grew in importance and essentiality (Oakes 1986). In other words, if an investigator uses such a statistic, then his findings are more likely to be acceptable to the

Barclay 26 gatekeepers of sociological knowledge and to find access to the sociological hierarchy via publication in prestigious journals (Weigert, 1970:116). In July of 1973, a man by the name of George Sternlieb, Professor of Urban and Regional Planning at Rutgers University, chose a noteworthy (to this thesis) phrase for an article title. His article was called The Sociology of Statistics: Measuring Substandard Housing. While the intent of the article was related to the United States Department of Housing and Urban Development, the fact remains that Sternlieb was one of the first to identify another aspect of sociology. He stated that what has been missing in the components of measurement is: a sociology of statistics: the why of what is being countedor sometimes even more interestinglynot counted.A macro-view of the institutions surrounding or absent in an area of data collection, rather than the internal view of the data content, can tell us things which though different from ordinary uses of the data collection results may be, at least in some regard, equally meaningful. (1)

WHAT IS RHETORIC IN STATISTICS According to Websters Dictionary, rhetoric is t he art or study of using language effectively and persuasively, or a skill in using language effectively and persuasively (1995:450). According to Plato, Socrates said it was an experience in creating a kind of pleasure and satisfaction (as cited in Cooper 1997). According to the opening chapter of The Rhetoric of the Human Sciences, rhetoric is not the ornamentation in a text or exploitation of words or deceptive writing (1987). With so many definitions, how can this broad idea be conceptualized to fit under the umbrella of statistics? The dilemma lies in that rhetoric is typically thought of as a tool of literary devices. To add one more source to the list, William Zinsser, in his book Writing to Learn, discusses the idea of written material on a variety of fields. He comments that often times people do not even

Barclay 27 realize there is literature on virtually every subject, that for example, mathematics consists of more than simply right or wrong responses to questions (1988). Through the process of mathematization, the study of mathematics can be classified as a science. However, this requires it to be thought of in a purely axiomatic and strictly numerical sense, void of all rhetorical device (Davis and Hersh 1987). This solution of scientizing mathematics is not accurate; no reason for eliminating verbal, or literary, context is valid. All subjects are reliant on words to express the intended meaning. Statistics contains an equal parallel. Statistics may not be as comprehensive a language as traditional wording, but nonetheless it is still a language. Arney describes language as a collection of symbols with a widely accepted set of rules for their manipulation (1979:173). Thereby it can be assumed that through the shared understanding with others, statistics, as a language, is a form of rhetoric. Symbols by themselves do not have meaning unless one is attributed to it. This can only be done by using words to describe the relationship it has with the number associated with it or the relationship it has to other symbols. Also interconnected with the idea of language is the idea that a word does not become a word unless it has a commonly shared interpretation and particular use; similarly, a statistic does not become a statistic unless it is agreed upon that it works and is important for something (Arney 1979). To conclude the proof that statistics is a form of rhetoric, or that it utilizes rhetoric in its tools for comprehension, Davis and Hersh make the following point (about mathematics, but in the given context the same can be thought of in terms of statistics): The myth of totally rigorous, totally formalized [statistics] is indeed a myth. [Statistics] in real life is a form of social interaction where proof is a complex of the formal and the informal, of calculations and casual comments, of convincing argument and appeals to the imagination and the intuition. The competent professional knows what are the crucial points of his argumentthe points where the audience should focus its skepticism. Those are the points where he will take care to supply sufficient detail. The rest of the

Barclay 28 proof will be abbreviated. This is not a matter of the authors laziness. On the contrary, to make a proof too detailed would be more damaging to its readability than to make it too brief. Complete [statistical] proof does not mean reduction to a computer program. Complete proof simply means proof in sufficient detail to convince the intended audiencea group of professionals with training and mode of thought comparable to that of the author. Consequently, our confidence in the correctness of our results is not absolute, not is it fundamentally different in kind from our confidence in our judgments of the realities of ordinary daily life. (1987:68)

IN SOCIOLOGY Within the field of sociology, a basic path exists to allow basic sociologists to be viewed in a generalized manner. This general model this path takes begins with the initial awareness of sociology as a collective sort of science (as seen through an introductory course); second is the traditional understanding of the forms research takes and the undercurrents permitting those who seek more information to proceed in a standardized manner (methods course); third, the necessary information regarding the theoretical perspectives outlining the material being studied and the greats in the field who have studied a similar phenomenon (theory); fourth, the acknowledgment that a barrier exists between the language of what is desired to obtain and what is permissible or plausible within the context of the study (dilemma of operationalization and validity); fifth, and finally, the realization that what seemed a mystery at the onset of the research really is not a mystery, but can be broken down into basic sociological concepts where the statement made, in essence, forms the reality. To grasp the extent to which sociology is rhetorical, one must understand how this model is related to the practice of sociology. Insomuch as all these steps aid in the purpose of rhetoric, which is to identify oneself and to persuade others (Weigert 1970:113), they each provide a unique knowledge of the researcher (in how he or she interprets the results), as well as the subject matter. As these two insights simultaneous develop, the field of sociology is furthered.

Barclay 29 On another level (one that is much further developed in Weigerts The Immoral Rhetoric of Scientific Sociologyhere I only seek to note a point worth mentioning, but not necessary for the utility of the thesis), as the researchers conduct research, the question is posed as to whether they are completing the task at hand for their own knowledge and pursuit of truth for the goodness of the field, or if they are merely working off of their own innate desires to achieve more, publish more, and therefore attain more notorie ty. There need not be a total distinction between the two, but that the researchers ought to come to terms with the driving force behind their work. If this distinction is recognized, a separate set of reasons why sociology is a form of rhetoric is revealed (Weigert 1970). Another language associated with whether or not the researchers are motivated by self- interests links the sociological language to a more sophistic undercurrent. This, Plato says, is the distinction between the sophist, who seeks toward rightness, and the philosopher, who seeks toward truth (Plato, as cited in Cooper, 1997).

IN THE SOCIOLOGY OF STATISTICS Weigert gave the example of a professor who submitted an article for publication in a well-known journal. The article was declined, but the professor was told that if he were to include more empirical data, his chance at being published would increase. He included a table of data that was only tangentially related to the topic of the research and made his resubmission. The revised article was soon published (116). This is just one way in which the power of statistical evidence is manifested through the public eye. People, both statistical advocates and those unassociated with such matters, are influenced by the use of numbers to prove a point. As Danziger noted, Faith in this methodology certainly unites a much larger number of research psychologists that does any kind of commitment to a particular theoretical framework (as cited

Barclay 30 in John 1992:145). This faith is a double-edged sword in that it provides a concrete ground for which researchers can state their conclusions, but it is also often misinterpreted and therefore the audience is powerless to reinterpret the data unless it is readily available in a non-biased form. Statistics as rhetoric in sociology need not be a self-depreciating practice. Instead of seeing it as a barrier to be hurdled or a feat to correctly explain, it is most suitably described as a looking- glass. It magnifies a topic by focusing on one specific application of its usefulness and allows the audience to see and understand as many of the factions of its components as they dare to explore. Statistics opens the doors for others to follow, duplicate, or even reconfigure the works of those who preceded them. It allows the audience to see for themselves all the data with which the researcher used to come up with his or her conclusions in a way that is non-threatening and accessible to any willing to discover the established and the soon-to-be-uncovered. Statistics as rhetoric is a compelling tool in the hands of those prepared to take the challenge of correct usage. If in numbers lies proof, in the practice of rhetoric lies true power.

IN REGARDS TO THE NULL HYPOTHESIS Rhetoric is many things to many people; several examples of this incorporation into various fields (e.g. statistics, sociology, and the sociology of statistics) were mentioned above. Now that a greater grasp is held of what rhetoric is and how is applies to general fields, duly important to note is its relevance to an exact practice and interpretation. The area of statistics mentioned above is the null hypothesis. In the example, the results led to a test statistic (or z-score) of 1.55 and therefore a onesided p-value of 0.0606. This illustration was chosen specifically because it proved a point. Traditionally the cut-off for insignificance versus significance is 0.05. If the test statistics led to

Barclay 31 values above 0.05, then the data was insignificant, and thus, below 0.05, significant. The current obsession with .05, it would seem has the consequence of differentiating significant research findings and those best forgotten, published studies from unpublished ones, and renewal of grants from termination (Skipper, Guenther, and Nass 1967:16). Can such a point be the end all and be all of statistics? Is it really possible that all consequence falls on one number and its placement relative to 0.05? In this example, the data would conclude that the findings were not significant and therefore it would not be necessary to conduct further research. Is this good? Is this right? If the p-value was 0.0499 would we conclude that the results were significant, and so much more so than 0.0606? Skipper, Guenther, and Nass say there is a distinction between 0.05 and 0.06 that determines right from wrong, legitimate versus meaningless, or even good versus bad (1967). This is a matter of rhetorical determinance. There should be no universal limit set on the importance of an argument. Perhaps the leading figure in the creation and history of statistics, R.A. Fisher, is known for establishing this crucial 0.05 point. However, even as he wrote and explained the principles associated with statistics, he clearly stated that 0.05 should not be the unquestioned axis from which all statistical significance can be derived. The value was simply a guide to be used for inference (1973). The importance of the analysis and interpretation lies in how it is interpreted and what relevance it has to society. Statistical significance need not imply semantical significance (meaningfulness), much less truth (Orey, Garrison, Burton 1989:17). Since the results are only probabilities, they can tell their readers nothing about how heeding the results from this collected data can impact society (Gorard 2003). Numbers do not have meaning inherent to them. The meaning is only placed when one establishes an interpretation to the numbers and explains itverballyin a way that makes sense to the audience.

Barclay 32

ACCEPTANCE V. FAILURE TO R EJECT The term null hypothesis in itself does not have meaning attributed to it either, unless the audience gives it meaning. One understands the null hypothesis significance test to be a test of whether or not there is a probability of getting a test statistic as big as is found when there is no significant difference between the two samples. If this were a small probability, one would agree that the results show a rejection of the null hypothesis. (For the moment) 0.025 would produce a rejection of any null hypothesis test and the acceptance of the alternative hypothesis. However, can the reverse of this be true? Can one say that if the p-value is greater than 0.05, at say 0.20, that the null hypothesis is accepted and the alternative hypothesis is rejected? Logically this would make sense. If the rejection of A proves B is accepted, wouldnt saying the rejection of B proves A is accepted? Yet, this is not the case (See FIGURE 2). Some may argue there is no difference in the statements but this is incorrect. The null hypothesis, one of the tricks of the trade if you will, cannot be proven true. More and more evidence can be built up for it, but no point exists where the results acquiesce to its acceptance (Berk 2001). The best one can aim toward is the failure to reject. In traditional null hypothesis testing, the failure, or refusal, to correctly articulate this phrase is heresy against the study (Murphy and Myors 2004). The reoccurring theme of rhetoric once again is apparent. In order to properly convey the message that the difference between the two samples is not significant, one is only left with two options: acknowledge that the data is not significant to prove a difference (i.e. failure to reject the null) or gather more data to retest their analysis.

Barclay 33 FIGURE 2TO ACCEPT/TO R EJECT/TO FAIL TO R EJECT

If the rejection of A proves B is accepted , then does the rejection of B prove A is accepted? Nope, not here! In the same manner: the rejection of the null hypothesis proves the alternative hypothesis is accepted, but the rejection of the alternative hypothesis DOES NOT prove the null hypothesis is accepted.

WHERE DO WE GO FROM H ERE? Now that a precedent for understanding the null hypothesis has been set, is there a way to aid in the accuracy of others ability to comprehend all that is going on in the field as a practice and a process? The notion still stands that it is possible for statistics to become a way of misrepresenting the truth. One of Darrell Huffs books is even titled How to Lie With Statistics (1954), and one of Joel Bests, Damned Lies and Statistics: Untangling Numbers From the Media, Politicians, and Activists (2001). (As a side note, both texts make a strong attempt at progressing the field of a sociology of statistics, even without calling it such. As more and more literature enters mainstream and people encounter the topics, it is virtually impossible for the long-reaching threads of its impact to go unnoticedwhich is all the better for sociologists and inquirers like ourselves!) The point has come, however, where numbers and statistics need to be looked at in a synthesized manner. Statistics is in no way to be condemned for its malleability of data, but ought to be embraced for its worth and potential. In order to create a positive outlook on the evaluation of the null hypothesis, one can interpret the work of Max Weber and his pure, or ideal, type and the parallels with the study of the null hypothesis. Weber says: Sociologyis a science which attempts the interpretive understanding of social action in order thereby to arrive at a causal explanation of its course and effects. Action in this sense may be either overt or purely inward or subjective; it may consist of positive intervention in a situationAction is social in so far as, by virtue of the subjective

Barclay 34 meaning attached to it by the acting individualit takes account of the behaviour [sic] of others and is thereby in its course. (1964:88) In these terms, the study of the null hypothesis falls under the realm of sociology. It is not only in a statistical or mathematical world that it can be studied; the interpretations of the p-values have a distinct meaning attached to them, pushing them to the sociological realm. The action of interpretation takes into account behavior of others and how each individual will respond to the test of significance. Weber continues by discussing the pure type. This, he emphasizes, is not singular in its evaluation by others. No correct or true meaning is established besides one that corresponds to the given information at a give n time. It is not like other sciences where one value of x corresponds with one and only one value of y. Similarly the null hypothesis and p- value combination is an example of the pure type because it provides a model for clarity within inferential statistics. The null hypothesis takes data and synthesizes it so that a single number (the p-value) is representative of the entire interpretation. Weber points out that the pure type is theoretically conceived and is of subjective meaning attributedin a given type of action (1964:89). Stated alternatively, the pure type, which in this case is the combination of the null hypothesis and p-value, is created through a theoretical basis (ones evaluation of significance) and is subject to various analyses by each individual through the act of interpretation. The null hypothesis, while at times seen as confusing or subjective to ones own evaluation, offers great potential to understanding not only more about the population from which the sample was taken, but also a look into how the language of inferential statistics revolves around the understanding and analysis of this ideal type.

Barclay 35 QUANTITATIVE ANALYSIS: RESULTS : From the qualitative analysis the basics of the null hypothesis and its interpretation were analyzed for their content. This portion of the thesis looks at the applicability of these interpretations and how both students and professors in the fields of Economics, Management, Mathematics, Psychology, Sociology, and Statistics respond when surveyed about their statistical understanding. The departments of Economics, Management, and Mathematics were grouped together, as were the departments of Psychology and Sociology. The reason for this is because of the related tendency either toward statistical dependence or away from such a dependence (and often towards qualitative procedures) the within the fields. When the department of Statistics is tested, those respondents fall into the category of statistical dependency. The findings of the survey are exp lained in the text of this analysis and any conclusions drawn from the analysis will be explored for further inference. The first table of analysis deals with the difference in attitudes toward statistics. One question on the survey asked for the students level of enjoyment of their statistics course. The following (TABLE 3) is a breakdown of the opinions of the five majors surveyed only.

Barclay 36 TABLE 3WHETHER OR NOT STUDENTS (BY D EPARTMENT) ENJOY S TATISTICS Strongly Agree/ Neutral Agree Econ/Mgmt/Math Psych/Soc Total 23 9 32 9 10 19 Strongly Disagree 4 18 22 36 37 73 Disagree/ Total

35 30 25 20 15 10 5 0 Strongly Agree/Agree Neutral Disagree/Strongly Disagree Econ/Mgmt/Math Psych/Soc

Null Hypothesis: The scores of enjoyment levels would be the same across all related majors.

Alternative Hypothesis: There is a difference in the scores of enjoyment levels across the related majors.

By running a chi-square analysis test, we see that the observed values are quite different than that of the expected values. With 2 degrees of freedom (d.f.), we have a chi-square value of 15.08. This corresponds with a p- value of 0.001. With a p-value well below the 0.05 statistical level, the data shows a significant relationship between the two variables. In other words, there is a link between whether or not a student likes statistics and the major he or she has chosen. For possible reasons to be discussed later, those majoring in the fields of economics, management,

Barclay 37 and mathematics tend to enjoy their study of statistics more than those in the fields of psychology and sociology. By performing a chi-square test, further interpretations of causal explanation cannot be conferred, but with a link as strong as this one, a more in-depth look at the relationship would be worthwhile (and will be addressed shortly). The next table (TABLE 4) looks at whether or not students in the five majors would take a statistics course even if it was not required.

TABLE 4WHETHER OR NOT STUDENTS (BY D EPARTMENT) WOULD TAKE A S TATISTICS COURSE IF IT WAS NOT REQUIRED Strongly Agree/ Neutral Agree Econ/Mgmt/Math Psych/Soc Total 15 5 20 8 4 12 Strongly Disagree 13 28 41 36 37 73 Disagree/ Total

45 40 35 30 25 20 15 10 5 0 Strongly Agree/Agree Neutral Disagree/Strongly Disagree

Econ/Mgmt/Math Psych/Soc

Barclay 38 Null Hypothesis: The scores of whether or not a student would take a statistics course if it was not required would be the same across all majors. Alternative Hypothesis: There is a difference in the scores of whether or not a student would take a statistics course if it was not required across the majors.

By running a chi-square analysis test, we see that the observed values, once again, are quite different than that of the expected values. With 2 d.f., we have a chi-square value of 11.81. This corresponds with a p- value of 0.003. Once again, one sees that a relationship between whether or not a student would take a statistics course eve n if it was not required and his or her choice for a major. Students in the fields of economics, management, and mathematics tend to be more willing to take a statistics course even if it were not required of them. Students in psychology and sociology are less likely to still want to take the course. The next table (TABLE 5) relates to whether or not statistics makes students nervous.

TABLE 5WHETHER OR NOT STATISTICS M AKES STUDENTS NERVOUS (BY D EPARTMENT) Strongly Agree/ Neutral Agree Econ/Mgmt/Math Psych/Soc Total 6 18 24 7 8 15 Strongly Disagree 23 11 34 36 37 73 Disagree/ Total

Barclay 39

40 35 30 25 20 15 10 5 0 Strongly Agree/Agree Neutral Disagree/Strongly Disagree

Econ/Mgmt/Math Psych/Soc

Null Hypothesis: Across the majors, the scores of whether or not a student statistics makes a student nervous would be the same.

Alternative Hypothesis: There is a difference across the majors in the scores of whether or not statistics makes a student nervous.

By running a chi-square analysis test, we see that the observed values, yet again, are quite different than that of the expected values. With 2 d.f., we have a chi-square value of 10.29. This corresponds with a p-value of 0.006. This number shows that there is, again, some relationship between students majors and their statistical anxiety. Students in economics, management, and mathematics are less likely to be nervous about statistics than those in psychology and sociology. From the breakdown of those who took the survey the following table (TABLE 6) is based on the knowledge of both students and professors in all six fields (Economics, Management, Mathematics, Psychology, Sociology, and Statistics).

Barclay 40 TABLE 6FAMILIARITY WITH THE N ULL H YPOTHESIS Yes Have you heard of the null hypothesis? 82 (95.35%) Have you used the null hypothesis in statistical analysis of data? 69 No 4 (4.65%) 16

(81.18%) (18.82%)

From this table, one sees the relevance of using statistical analysis of data. While looking at the results, reader, keep in mind that every one of those who has completed this part of the survey has taken (or is currently taking) a statistics course. Since the survey was not administered until well into the semester, all statistical courses have incorporated the null hypothesis into the curriculum at this point. The following is a table describing both students and professors views on the null hypothesis, its usage, and understanding (TABLE 7).

Barclay 41 TABLE 7VIEWS ON THE N ULL H YPOTHESIS Strongly Agree I understand the null hypothesis (eg. setup, usage, conclusion). When I encounter a question in the text dealing with the null hypothesis, I feel confident I can analyze the data. I understand how the null hypothesis is important to the analysis of research. I understand what it means for data to be statistically significant. 39 (45.35%) 42 (48.84%) 3 (3.49%) 2 (2.33%) 0 (0.00%) 20 (23.53%) 46 (54.12%) 8 (9.41%) 9 (10.59%) 2 (2.35%) 20 (23.26%) 14 (16.28%) 42 17 7 (8.14%) 17 Agree Neutral Disagree Strongly Disagree 0 (0.00%) 1 (1.16%)

(48.84%) (19.77%) 38 16

(44.19%) (18.60%) (19.77%)

This data is simply a background display to show that students and professors alike are fairly certain of their statistical knowledge. The majority of respondents agreed or strongly agreed with the statements. Now that all the background data referring to the students and professors opinions has been charted, the information regarding the individual questions on the interpretation of the null hypothesis will be presented. The following tables (TABLES 8-14) visually show the breakdown of interpretations between the six departments. For each table the null/alternative hypothesis is the same:

Barclay 42 Null Hypothesis: There is no difference across the departments in ones answer to the question. Alternative Hypothesis: There is a difference across the departments in ones answer to the question. What follows each table is the chi-square analysis: the corresponding d.f., chi-square value, and p-value, as well as the correct answer, a brief, straight- forward interpretation of the results, and a pie chart showing the correct answer versus the incorrect.

TABLE 8R ESULTS TO Q UESTION 1 Question: The _________ hypothesis says that the difference between the samples is due to chance and nothing else. Alternative Econ/Mgmt/Math/Stat Psych/Soc Total 25 14 39 Null 34 36 70

On 1 d.f., the chi-square value is 2.43, which corresponds with a p-value of 0.119. Correct Answer: Null Hypothesis The results show there is not a significant link between the field of study or instruction of an individual and the answer each chose to this question. However, the pie chart below shows that 67% of the respondents answered correctly.

Barclay 43

TABLE 9-- R ESULTS TO Q UESTION 2 Question: Given all other things to be equal, what p-value gives the most evidence for the null hypothesis. 0.01 Econ/Mgmt /Math/Stat Psych/Soc Total 22 20 42 0.10 0 1 1 0.50 7 7 14 0.99 30 9 39

On 3 d.f., the chi-square value is 7.77, which corresponds with a p-value of 0.051. Correct Answer: 0.99 The results show there is a link between the field of study or instruction of an individua l and the answer each chose to this question. This is a prime example of a situation where the pvalue is near the traditional cut-off point of 0.05. In such cases it is helpful to look at the question and the responses, as well as the sample size. At all times one must keep in mind

Barclay 44 Fishers principle (1973) that serves as a reminder that no point should be set that must determine significance in all circumstances. For this instance, as will be later discussed, the issue changes from becoming what is technically significant to what must be gleaned from its analysis and introspection. An important note, however, is that the pie chart below shows only 42% of the respondents answered correctly.

TABLE 10-- R ESULTS TO Q UESTION 3 Question: A p-value is the chance of getting a big test statistics if the null hypothesis is correct. True Econ/Mgmt/Math/Stat Psych/Soc Total 20 16 36 False 34 27 61

Barclay 45

On 1 d.f., the chi-square value is 0.00, which corresponds with a p-value of .99. Correct Answer: True The results show there is not a significant link between the field of study or instruction of an individual and the answer each chose to this question. However, the pie chart below shows that only 38% of the respondents answered correctly.

TABLE 11-- R ESULTS TO Q UESTION 4 Question: If the p- value is less than 0.05, the result is __________________. Statistically Significant Econ/Mgmt/Math/Stat Psych/Soc Total 37 33 70 Statistically Insignificant 17 12 29

On 1 d.f., the chi-square value is 0.27, which corresponds with a p-value of 0.60.

Barclay 46 Correct Answer: Statistically Significant The results show there is not a significant link between the field of study or instruction of an individual and the answer each chose to this question. However, the pie chart below shows that 69% of the respondents answered correctly.

TABLE 12-- R ESULTS TO Q UESTION 5 Question: If the p- value is 0.25, it is best to say that the null hypothesis is ______________. Accepted Econ/Mgmt/Math/Stat Psych/Soc Total 7 11 18 Not Rejected 34 21 55 Rejected 17 13 30

On 2 d.f., the chi-square value is 2.90, which corresponds with a p-value of 0.23. Correct Answer: Not Rejected

Barclay 47 The results show there is not a significant link between the field of study or instruction of an individual and the answer each chose to this question. However, the pie chart below shows that 54% of the respondents answered correctly.

TABLE 13-- R ESULTS TO Q UESTION 6 Question: Can a p-value of 0.06 ever be considered statistically significant? Yes Econ/Mgmt/Math/Stat Psych/Soc Total 47 32 79 No 12 12 24

On 1 d.f., the chi-square value is 0.68, which corresponds with a p-value of 0.41. Correct Answer: Yes

Barclay 48 The results show there is not a significant link between the field of study or instruction of an individual and the answer each chose to this question. However, the pie chart below shows that 78% of the respondents answered correctly.

TABLE 14-- R ESULTS TO Q UESTION 7 Question: Can a p-value of 0.25 ever be considered statistically significant? Yes Econ/Mgmt/Math/Stat Psych/Soc Total 32 19 51 No 27 26 53

On 1 d.f., the chi-square value is 1.47, which corresponds with a p-value of 0.22. Correct Answer: No

Barclay 49 The results show there is not a significant link between the field of study or instruction of an individual and the answer each chose to this question. However, the pie chart below shows that only 47% of the respondents answered correctly.

ANALYSIS : Based on the results from the previous section, one can see the relationship (or lack thereof) between different understandings and interpretations of the null hypothesis across the fields of Economics, Management, Mathematics, and Statistics and then of Psychology and Sociology. From the first set of tables, a distinction is evident based on the feelings relative to ones enjoyment of statistics, likelihood to taking a statistics course if it was not required, and the anxiety associated with statistics. In all three areas a tendency exists between the three fields of economics, management, and mathematics to show similar evidence as one another and then for psychology and sociology to show alternative evidence. This being true, it makes one pause to wonder if there is a correlation between the department in which an individual takes a statistics course and the relevance statistics has in his or her particular field. In other words, does the

Barclay 50 ability of the classroom activities or instruction to relate to the specific field of interest to the student play a role in determining whether or not he or she likes and feels comfortable with the information? Given that the business, mathematics, and psychology departments all offer statistics courses within the field, is it possible that a student in the field of economics, management, mathematics, feels more confident and enjoys statistics more than his or her counterpart in psychology or sociology? Lundgren and Fawcett argue that sociology students should only take statistics courses within the sociology department. Due to several factors, different methods, different levels of measurements, different sample sizes, and different statistics, students do not see the relevancy to the course material in other departments (1980:195). The second group of tables relates to ones comprehension of the null hypothesis and his or her confidence in relaying the material and acknowledging the relevance of statistical knowledge to research. As mentioned before, they were used as a tool for the reader to better recognize the levels at which the sample saw itself as a demographic. Interestingly enough, however, several students said they had never heard of the null hypothesis or had yet to use it in statistical analysis of data. Given their status as students of at least one statistics course, one would wonder how it is possible they have not heard or used the tests in their research. In all the last seven tables above, one will notice that the p-values for each of the chisquare analysis tests are insignificant (with one being on the border of significance) by the standards of conventional analysis.

Barclay 51 CAVEAT IN THE FINDINGS : This is not to say that the testing was useless or insignificant to research. In MacRaes article, Measurement Scales and Statistics: What can significance tests tell us about the world?, he explains that there are no wrong questions in statistics, only wrong interpretations. To state it differently, the results of the significance test do not show that the data is insignificant, only that what has been proven is altered from what was originally sought. While the data does not show a significant difference across majors, minor, and instructors in various fields, it does show a lack of correct interpretations related to the world of statistics. In the last seven questions a range of answer percentages emerged from the respondents. In a given question, the results ranged from being 78% correct, all the way down to being only 38% correct. Through data manipulations and inversions, the results of this survey could be altered so as to obtain numbers significant to modern standards. While this may be helpful to the point sought in this text, it is not relevant or ethical to show anything less than the truth. The data did not correspond with what I had hoped to find, yet is still quite significant to the field. The very essence of this survey shows how important it is to have a complete and comprehensive grasp on the rhetoric of statistical analysis and tests. The results were disappointing, statistically speaking, to a researcher, but still comparable to topics of study in modern research. Only one of the questions analyzed yielded a p-value that showed a possible response worth studying and interpreting within its context. What is helpful, however, is the acknowledgement that more needs to be done in the social sciences, in particular, to explicate the usefulness of applying rhetorical skills to inferential statistics to the various fields of study. Whether or not the students have a different grasp on the knowledge across the departments, one thing is certain from this study: students who learn statistics relative to their

Barclay 52 field are more likely to find enjoyment in their statistics classes and be less nervous about the material. Garfield (as cited in Albert) stated that a general principle in learning statistics is that students learn by constructing their own meaning for what they are learning (2003:37). When students do not see the relevance between their topic and the numbers they are analyzing, they tend to overlook that statistics can be a help to their future studies and they see it as a barrier blocking them from obtaining their degree (Onwuegbuzie and Leech 2003).

Q UANTITATIVE CONCLUSIONS : In as much as Weber and his depiction of the ideal type when studying the null hypothesis applies, so do the numbers themselves apply as action that is social. The passages into inferential statistics apply using the same connection. Using another quote from The Theory of Social and Economic Organization, Weber states, The more sharply and precisely the ideal type has been constructed, thus the more abstract and unrealistic in this sense it is, the better it is able to perform its methodological functions in formulating the clarification of terminology, and in the formulation of classifications, and of hypotheses (1964:111). In order to relate the doing of the null hypothesis to the ideal type, one must understand how it is viewed by the larger audience. From this survey one sees how the more carefully the analysis is performed and the more precise an interpretation is required, the more abstract and unrealistic it is to acquire the results and interpretations one typically expects. The doing of the null hypothesis is, therefore, the process of correctly articulating the rhetoric of its evaluation. In turn, the more abstract the results are in relation to the general understanding, the more one is able to better explain the terminology and the corresponding hypothesis of its usage.

Barclay 53 OVERALL CONCLUSION: Statistical methods are supposed to help people make wise decisions, based on reasonable analyses of reliable data. But even after some training in statistics, many people dont understand what is required to obtain reliable data, they dont know how to determine if a reasonable analysis has been done, and they dont know how to make wise decisions based on the results (Utts 2003:73). If statistics in the classroom (and on university surveys) is a source of anxiety and angst among students, something ought to be done toward finding a solution. A great importance ought to be placed on working toward a succinct understanding and full knowledge of not only the doing, but the rhetoric of the null hypothesis. Functional numeracy, a concept defined by Eric Sowey, is a competence to understand, and argue back to, a statistic and to draw accurate meaning from, and criticize, a statistical argument about the real world (2003:89). The general public has a fear of statistics and the ability to analyze numbers; students are no different, nor are students of sociology. Within the field, one must be careful not to curtail the amount of statistical information offered to the student because of a fear of misinterpretation. There ought to be more courses designed to accentuate the detail and evaluation of inferential statistics in the social sciences. In the last study conducted on the subject, only about thirty percent of the undergraduate sociology programs [in 1966] in the United States offered a course in statistics (Reid and Bates, as cited in Shin 1975:93). Because of this, Schacht and Stewart claim that more students are turning to the social sciences simply to avoid mathematical calculations and statistical analysis. Since most sociology requirements do not include many courses in the field, more anxiety is placed on the doing of statistics, which causes the students to avoid them even further because of their uncertainty in their abilities to perform the analysis (1990). Statistics ought to be taught like the language it is (Lazar, as cited in Forte, 1995). The fact that many people are hesitant to perform calculations does not make it all right to ignore.

Barclay 54 Statistics plays a key role in both the classroom and society, in general. Weber urges the sociologist to create a pure type with the available forms of action to ensure the highest level of the logistical assimilation of meaning. However, this makes such a creation even harder to possess (1964). While the potential for establishing an ideal type so pure that all things going against it are dismantled, one should still press on toward its conception. Statistics as a form of rhetoric in the social sciences is essential to the understanding of society. We ought to treat it with respect. While we may not go as far as to say that all we know about a person we learn from statistics, to treat the field as a cold and distant relation will not grant us further knowledge. We must accept that statistics are part of our everyday lives, learn the language, and never cease to explore its wealth of knowledge.

Barclay 55 References

Abelson, Robert P. 1995. Statistics As Principled Argument. Hillsdale, NJ: Lawrence Erlbaum Associates, Publishers. Albert, James H. 2003. College Students Conceptions of Probability. The American Statistician 57 (1): 37-45. Arney, William R. 1979. Statistics as Language. Teaching Sociology 6 (2): 173-178. Baudrillard, Jean. 1999. Simulacra and Simulations: Disneyland. Pp. 481-486 in Social Theory, edited by Charles Lemert. Boulder, CO: Westview Press. Berk, Richard A. 1987. How Applied Research Can Save Sociologists From Themselves. The American Sociologist 18(2):119-126. Best, Joel. 2001. Damned Lies and Statistics: Untangling Numbers From the Media, Politicians, and Activists. Berkeley, CA: University of California Press. Coolidge, Frederick L. 2002. Statistics: A Gentle Introduction. Thousand Oaks, CA: Sage Publications. Cooper, John M., ed. 1997. Complete Works of Plato. Indianapolis, IN: Hackett Publishers. Creswell, John W. 2003. Research Methods, 2nd Ed. Thousand Oaks, CA: Sage Publications. Dar, Reuven. 1987. Another Look at Meehl, Lakatos, and the Scientific Practices of Psychologists. American Psychologist Feb 1987: 145-151. Davis, Philip and Reuben Hersh. 1987. Rhetoric and Mathematics. Pp. 53-68 in The

Barclay 56 Rhetoric of the Human Sciences: Language and Argument in Scholarship and Public Affairs, edited by John Nelson, Allan Megill, and Donald McCloskey. Madison, WI: The University of Wisconsin Press. Durkheim, Emile. 1965. Elementary Forms of the Religious Life, translated by J.W. Swain. New York: Free Press. Fisher, R.A. 1973. Statistical Methods and Scientific Inference. New York: Hafner Press. Forte, James. 1995. Teaching Statistics Without Sadistics. Journal of Social Work Education 31(2): 204-218. Freedman, David, Robert Pisani, and Roger Purves. 1998. Statistics, 3rd Ed. New York: W. W. Norton & Company. Gal, Iddo. 2003. Teaching for Statistical Literacy and Services of Statistics Agencies. The American Statistician 57(2): 80-84. Gorard, Stephen. 2003. Understanding Probabilities and Re-Considering Traditional Research Training. Sociological Research Online 8(1): 1.1-7.2. Hooke, R. N.d. Humorous Quotes About Statistics. Las Vegas, NV: I-Laugh, Retrieved September 18, 2003 (http://www.workinghumor.com/quotes/statistics.shtml). Huff, Darrell. 1954. How to Lie With Statistics. New York: W. W. Norton & Company. John, I. D.1992. Statistics as Rhetoric in Psychology. Australian Psychologist 27 (3): 144-149. Lundgren, Terry D. and Richard Fawcett. 1980. "Statistics From Statisticians." Teaching Sociology 7 (2):191-201. MacRae, A.W. 1988. Measurement Scales and Statistics: What Can Significance Tests Tell Us About the World. The British Psychological Society 79: 161-171. McPherson, Glen. 2001. Applying and Interpreting Statistics: A Comprehensive Guide, 2nd Ed.

Barclay 57 New York: Springer-Verlag. Murphy, Kevin R. and Brett Myors. 2004. Statistical Power Analysis: A Simple and General Model for Traditional and Modern Hypothesis Tests, 2nd Ed. Mahwah, NJ: Lawrence Erlbaum Associates, Publishers. Nelson, John, Allan Megill, and Donald McCloskey. 1987. Rhetoric of Inquiry. Pp. 3-18 in The Rhetoric of the Human Sciences: Language and Argument in Scholarship and Public Affairs, edited by John Nelson, Allan Megill, and Donald McCloskey. Madison, WI: The University of Wisconsin Press. Oakes, Michael W. 1986. Statistical Inference: A Commentary for the Social and Behavioural Sciences. Chichester, NY: Wiley. Onwuegbuzie, Anthony, J. and Nancy L. Leech. 2003. Assessment in Statistics Courses: more than a tool for evaluation. Assessment & Evaluation in Higher Education 28 (2): 115127. Orey, Michael A. III, James W. Garrison, and John K. Burton. 1989. A Philosophical Critique of Null Hypothesis Testing. Journal of Research and Development in Education 22 (3): 12-21. Paulos, John A. 1988. Innumeracy: Mathematical Illiteracy and Its Consequences. New York: Vintage Books. ------. 1995. A Mathematician Reads the Newspaper. New York: Doubleday. Raftery, Adrian. 2001. Statistics in Sociology, 1950-2000: A Selective Review. Sociological Methodology 31(1):1-45. Rossman, Allan, J., Beth L. Chance, and Robin H. Lock. 2001. Workshop Statistics: Discovery With Data and Fathom. Emeryville, CA: Key College Publishing.

Barclay 58 Russell, Bertrand. 1959. Knowledge by Acquaintance and Knowledge by Description. Pp.4659 in The Problems of Philosophy. New York: Oxford University Press. Salkind, Neil J. 2000. Statistics for People Who (Think They) Hate Statistics. Thousand Oaks, CA: Sage Publications. Schacht, Steven and Brad J. Stewart. 1990. Whats Funny About Statistics? A Technique for Reducing Student Anxiety. Teaching Sociology 18 (January): 52-56. Shin, Eui H. 1975. Statistics Requirements for Undergraduate Sociology Majors in Colleges And Universities in the United States. The American Sociologist 10 (May): 92-102. Skipper, James K. Jr., Anthony L. Guenther, and Gilbert Nass. 1967. The Sacredness of .05: A Note Concerning the Uses of Statistical Levels of Significance in Social Science. The American Sociologist 2: 16-18. Sowey, Eric R. 2003. The Getting of Wisdom: Educating Statisticians to Enhance Their Clients Numeracy. The American Statistician 57(2): 89-93. Sternlieb, George. 1973. The Sociology of Statistics: Measuring Substandard Housing. Public Data Use 1 (3): 1-6. Turner, Stephen P. 1987. "Underdetermination and the promise of statistical sociology." Sociological Theory 5 (2):172-184. Utts, Jessica. 2003. Special Section on Statistical Literacy: How You Can Help. The American Statistician 57 (2): 73-79. Weber, Max. 1964. The Fundamental Concepts of Sociology. Pp. 87-115 in Max Weber: The Theory of Social and Economic Organization, translated by A.M. Henderson and Talcott Parsons. New York: The Free Press. Websters New American Dictionary. 1995. Editors of Merriam-Webster. Pp. 450. New York:

Barclay 59 Merriam-Webster, Inc. Weigert, Andrew J. 1970. The Immoral Rhetoric of Scientific Sociology. The American Sociologist 5: 111-119. Zinsser, William. 1988. Writing to Learn. New York: Harper & Row, Publishers.