Using Multivariate Statistics - Linda Fidell, Barbara Tabachnick

...all theory or all practice. Using Multivariate Statistics is a carefully crafted blend of pure theory, application and discussion which the authors update every five years or so. The material is revised regularly to ensure that the commentary on the computer packages is up to date and that any new fads or fashions in the world of research are included (eg structural equation modelling was added to the Third Edition after an upsurge the use of this ...
...on.

Using Multivariate Statistics is a moderately large tome (about 1 000 pages) covering multiple regression, canonical correlation, multiway frequency analysis, analysis of covariance, multivariate analysis of variance, profile analysis, discriminant function analysis, logistic regression, principal components/factor analysis, and the general linear model. The three appendices contain an introduction to matrix algebra, complete ... more

Most advanced statistics books are usually either all theory or all practice. Using Multivariate Statistics is a carefully crafted blend of pure theory, application and discussion which the authors update every five years or so. The material is revised regularly to ensure that the commentary on the computer packages is up to date and that any new fads or fashions in the world of research are included (eg structural equation modelling was added to the Third Edition after an upsurge the use of this technique).

If you’re already starting to feel a bit nervous or statistically inadequate you might want to skip the next couple of paragraphs and go straight to comments under the subheadings. The last thing I want to do it put anyone off what is a very accessible volume that can be easily understood by graduate students and professionals alike. Those who are brave enough for a brief overview of the contents take a deep breath and read on.

Using Multivariate Statistics is a moderately large tome (about 1 000 pages) covering multiple regression, canonical correlation, multiway frequency analysis, analysis of covariance, multivariate analysis of variance, profile analysis, discriminant function analysis, logistic regression, principal components/factor analysis, and the general linear model. The three appendices contain an introduction to matrix algebra, complete worked examples with their research designs and the (ever heart stopping) statistical tables to assist with making decisions about levels of significance. Tucked in the back is a data disk for those who like to see for themselves that the answer is actually 15.67 (I haven’t used the disk – I usually have my own data to work with).

Each chapter is broken down into comprehensive, clear sections that coax a novice reader without sacrificing the detail required by a more experienced researcher. Basic principals are explained, examples introduced and an applied cases are worked through for every technique. The authors are very conservative and devote quite a bit of attention to discussing the limitations and appropriate use of each approach. They have prevented me from applying an incorrect analysis or making cavalier statements on more than one occasion.

Why do I have a soft spot in my heart for Using Multivariate Statistics? My top five reasons are below.

Commonsense guidance and advice

The authors have managed to avoid any trace of academic elitism and limited the jargon to an absolute minimum. They’ve also discussed the most popular and widely used tools instead of looking at research from a statistical purists point of view. Commonsense chapters that explain how the book is organised and the best way to remove anomalies from data sets are extremely useful topics that most social science methods texts don’t bother to cover. Checklists summarising the steps associated with various processes are a helpful addition.

Comparison of packages

I use SPSS (Statistical Package for the Social Sciences) and have regular wars with SAS devotees. Using Multivariate Statistics cuts through ideological and philosophical differences by covering the pros and cons of all the major software ‘players’ (SPSS, SAS, SYSTAT and BMDP). Handy charts tell you unambiguously what each of the packages can and can’t do before you get too excited about a technique that your software may not be able to cope with.

Interpretation of statistical package output (usually SAS or SPSS)

The output interpretation pages are a huge strength of Using Multivariate Statistics. Usually, statistics textbooks vaguely refer to examining, for example, your R-squared residuals. The authors of this fine work include the actual print out from a software package, show you where the R-squared residuals are and discuss what the values mean in the context of the example. Unbelievably helpful when you’re faced by pages and pages of results and have no idea where to start looking for the answer.

Clear summary tables

Not sure which technique to use? Don’t worry: there are clear summary tables that carefully detail the circumstances and conditions for using each of the analysis methods.

Girl power!

I must admit I’m quietly pleased that the authors of such a magnificent technical work are both women (Barbara Tabachnick and Linda Fidell). During my last few years at school I was informed that I wasn’t capable of doing higher mathematics and had to stay in a less advanced class. My teacher even offered to write a letter home informing my parents of my lack of ability on the numeric front. Despite this self esteem boost (thanks for your support) I went on to complete a science degree with a statistics/research methods major.

And finally, the million dollar question, who should read Using Multivariate Statistics?

The simple answer is anyone with an understanding of basic statistical techniques who either wants to try or is currently undertaking complex research that goes beyond straightforward experimental designs. Using Multivariate Statistics is suitable for those working or studying in the social or natural sciences as well as people in business or marketing-related professions.

Oh, it also helps if you have a sense of humour: the authors most certainly do. Constant references to belly dancers (with a pixellated belly dancer on the cover, no less) make you smile even though your brain is hurting.

I can honestly say (almost without giggling)if you only buy one advanced statistics book in your life, it should be Using Multivariate Statistics.

(Oh, I went home and compared the last two editions - yes, I have both. The latest has two extra chapters on time series and survival analysis. Apologies to users of BMDP but you no longer rate a mention. Supporting data sets are now on-line.)

 

...less