Customer Reviews

Your Statistical Consultant: Answers to Your Data Analysis Questions

5 star:		(5)
4 star:		(2)
3 star:		(0)
2 star:		(1)
1 star:		(0)

 

 

As social science researchers we want to know three things:

1. Can we detect a relationship?

2. Can we define the strength of the relationship? and

3. Can we cognitively make sense of the relationship?

Other authors touch on Statistical Power and leave you hanging. Newton and Rudestam open your eyes to something going on out there in the research world that is very leading edge. They expanded me statistically and brought me through a place I've NEVER been before. I think I've got a total of about 7 hours into it now!

We, social scientists love to focus on the Type I error...its like Type I diabetes, an epidemic! Newton and Rudestam (p. 83) say, "A finding of statistical signficance means only that the true population effect is probably NOT zero, but typically it is misunderstood as indicating that the study is of substantive significance. A finding that a study is not statistically significant means that we have insufficient evidence to conclude that an effect is present, but it is often misunderstood to mean that the absence of an effect has been demonstrated." Read that about 7 times until you get it and you can call yourself a statistician!

Newton and Rudestam (Chapter 4) go into some really DETAILED explanations of the issues of researchers being enamored with and stopping with simple Type I error analysis. They define a series of studies (p.71) and say, "Sadly, the literature offers copius evidence of how research studies in the social sciences are underpowered for detecting all but very large effects." In order to get to a good statistical power level of .8 or higher you have to have some big samples sometimes. It depends upon the effect size of the population itself which you are modelling.

Bottom line is this "One cannot conclude the findings are large or important based on the significance level." (Newton, Rudestam, p.88). What they go on to explain very well is that you have to know the power of your testing, you have to estimate the effect size, you have to know if you sample size gives you the power you need. If you don't, you are one of the social scientists who are apparently so enamored with Type I testing that, "...rejecting a null hypothesis is akin to rejecting the proposition that the moon is made of green cheese." (p. 87)

As you can tell, I needed this book for its treatise on Power. But, you will find a great deal of other gems in this book. The authors have made it a "one stop shop" for statistical questions. If you have had a bit more than a basic stat class and need to think through stats for research methods so that you won't be seen as an idiot by your scholarly community, BUT THIS BOOK and dog ear it. I am doing just that with it.

I'm doing research in second language acquisition for my dissertation and have been struggling with everything related to statistics. It seems like every time I answer one question, three more pop up in its place. The statisticians I consulted with on campus were of no help because they couldn't translate their expertise into language that non-experts could understand, or they weren't familiar with more complex methods like general linear models. And I can only pester my dissertation advisor with stupid questions so many times. :) This book answers clearly and definitively almost all of the questions that I have posed to real, live statisticians and have not received satisfactory answers to. The title doesn't lie: it really is like your own, 24/7, personal consultant.

Some of the features that make this book so handy include a step-by-step worksheet/matrix for choosing statistical tests, followed by explanations of the assumptions and logic for each test plus issues in interpreting the results. The variety of clearly presented examples help to conceptualize difficult ideas and apply them to your own research. I'm almost 100% math-impaired, so I was stunned to understand the mathematical assumptions for each test that were described in the book. It also made clear for me for the first time how to interpret some of the graphs, plots and output charts that have confounded me in the past.

Before buying this book I was concerned that even it might be over my head because my exposure to statistics is limited to a two-week survey in a research methods class. However, the clarity and depth of the explanations in this book were just what I needed to deepen my understanding and allow me to use what I learned in my own research. I especially appreciate that the book concentrates on the most common questions that the authors have encountered about each test instead of providing too much information that serves no practical purpose.

My only complaint is that the book doesn't cover more "cutting-edge" (at least in my field) and complex tests like structural equation modeling (oddly, there is an entry for this topic in the index but no page number, and it is missing from the text). However, it covers everything that most grad student researchers will need to know about statistical tests.

If you are an educator or social scientist looking for guidance on applying the type of information contained in this book to anlayses using SPSS software, I also recommend "Doing Quantitative Research in Education with SPSS" by Daniel Muijs.

I had Rae Newton for the graduate methods course(s) at CSU Fullerton (502A and 502B). I have used this book many times to answer all of my questions (of course, when Dr. Newton was not available.) In short, I think this book should be used in all methods/stats classes. It really gets to the point and does not waste too much time with all of the formulas. I think it is meant to be used as a guide for non-math/statistics people who are using statistical packages like SPSS. It makes statistics understandable.

I wish I had come across this book sooner! I have a modest statistics background, but unlike most other statistics books, this book really opened my eyes to the reasoning behind various common statistical tests.

If you haven't taken an introductory statistics course before, then this book is probably not for you, but if you have, chances are that you'll find this book really helpful when doing data analysis. I plan to refer to it often.

The spirit of the book is described on page 54: "The biggest challenge is to be able to recognize which types of analysis are most appropriate for a given situation, and how to interpret and apply the results." [Kachigan, 1986]

My comments:

- The book is well-written, with lots of examples that are pretty easy to follow.

- The authors promote the use of exploratory data analysis to reveal the meaning of data between pairs of variables (e.g., plot bivariate cases, examine shapes of distributions, do cross-tabulations and scatterplots).

- They discuss, in detail, data preparation (e.g., data coding for survey questions), descriptive vs. inference statistics, the relationship between statistical significance and substantial significance, alpha vs. beta, power, effect size, etc. For example, how large should my sample size be if I want to detect significance at a given alpha level?

- They make the case for confidence intervals being more useful than a test of significance, especially smaller confidence intervals.

- They evaluate the assumptions behind data that approximates a normal distribution, and give suggestions for how to handle non-normal data (e.g., via power transformations).

- They explain how to deal with outliers and missing values.

- They discuss the assumptions behind various statistical tests, and explain when some test isn't going to be useful.

- They provide trade-offs in the consideration of sample size: the costs associated with collecting data, including the researcher's own time, and the requirements of statistical testing (e.g., power analysis comparing the number of independent variables, alpha level, and desired power level with the sample size).

- In response to the question, "How do I select the appropriate statistical test?", they begin by asking you to carefully consider the question(s) that you're trying to address with your data. They provide checklists or flowcharts for selecting the appropriate statistical test (e.g., p. 140+) depending on the kind of dependent variable (continuous, discrete) and whether the independent variable is discrete or continuous, how many independent variables there are, whether you want to control for other variables (covariates), etc. The flowcharts provide several design frameworks for you to choose. For example, does your research question concern the degree or strength of a relationship between the dependent and independent variable(s)? Are there significant group differences that you want to investigate? Etc. You're directed to the appropriate test (e.g., bivariate correlation, multiple regression, one-way ANOVA or t test, contingency table analysis, cross-tabulation, loglinear models, ...)

- They discuss, in substantial detail, the two important families of statistical approaches: ANOVA (analysis of variance), and multiple regression.

- They show the "bigger picture" of statistical tests, namely, how the various tests relate to one another based on the number of independent and dependent variables and their level of measurement (discrete, continuous). For example, from more general to more specific: canonical correlation -> MANOVA -> factorial ANOVA -> one-way ANOVA -> t test -> cross-tabulation or 2x2 tables; canonical correlation -> discriminant function or logistic regression; canonical correlation -> multiple regression and correlation -> bivariate regression -> bivariate correlation -> point biserial correlation -> cross-tabulation or 2x2 tables; point biserial correlation <-> t test).

- Towards the end of the book, they spend a bunch of time on meta-analysis. Meta-analysis allows us to summarize, aggregate, and compare results across multiple studies, and includes power and effect size. The last part of the book is somewhat deeper than earlier on, but it will be a very useful reference should I get to that level of research.

Overall, this is a wonderful statistics book ... a keeper for many years to come!

This book is not useful at all to a beginning level student. You would have to have another book in addition to this one in order to be able to select the appropriate operations and complete them. In addition, the completed results shown are not presented in a way that is comprehensible to a beginning level student. All in all, this book is only useful for someone who already knows what to do and how to do it.

The book is handy and good reference for students of statistics and related disciplines. The downside perhaps, is lack of worked examples on most of the topics covered.

I bought this book because it was required for a course, but I plan to keep it. Easy to understand, well-written and clear explanations of various statistical measures and how to apply them, with good examples throughout.

Nice book for most engineers that need direction on which stat to use. I have it on my work desk and great when needed a few times a month.