=NastyData Unruly, Ill-Mannered Observations Can Ruin Your h a l y s i s GARY H.cM

Researchers conkonring their own data often find them when they are detected. The good news is that those data to be more unruly. ill-mannered. and iras- the standard statistical methods are remarkably robust cjbIe than the well-behave4 cooperative dara found in againsr some violations of she assumptions. However, rextbook examples. Irasaile data that slap us in the face there are some violations of those assumprionsthat can at lean gt our attention. More dangerous are those wreak havoc on one's analysis. In this chapter we give stealthy, sinister observations that can go underected advice about how to separate the benign from the killer and yet have a disproponionare and untoward effect violations, on our analyses. This chapter describes techniques ior mere are many ways in which one's data can bedetecting and ~ & n g those nasty data that otherwise mme nasty and in-mannered. Data recording and data entry errors have the potential for introducing whopcould ruin your analyses. ping outliers. We may believe w e are sampling horn what appears to us to be one population but which in SOURCE OF PROBLEMS fact is several distincr populations, one possibly much The mm~osedly"tried-and-true' statistical methods smaller than the others. Or nature may jun throw us can be remarkably h g l l e in some situations. In par- a curve ball now and then to keep dxe game imerestticular. those methods that minimize sums of squares ing. m whatever ways our dara become irascible. it is including the most common sta~idcalprocedures used important for us to detefl the potential problems. by social psydalogins such as r-tests, ArVOVAs, and Mow smtistial analyses are performed by compurregression - are very scnsltive to audicrs. Estimates of err, but unfortunately the standard output from many group means or rcgmsion paramcten can sometimes programsjust provides the mud tesktatiscics. intermeof those test statisria, and he dramatically affected by just, a single outlier obser- diate steps in the c r a l d a ~ m vation. Moa textbooks from tvhich social pryrhologins their associated probabilities mder the null hypo^Iearn lheir statistics indicate that the standard proce- esis. Such outputs seldom provide durs abour n a q dures assume normality and homogeneity of variance data that mjght be lurking in the analysis. However. (equal variances within each group or equal variances almost all computer packages also now provide adearound each predicted criterion in regression). Tables quate graphical and statistical procedures for d e t e h g of meal values or those 'p-values' produced by sta- D U and data ~ that ~ violate the standard asmmptions. risical prografns as well as statidcat power analysis h n d the rekvant procedures are mostly graphical and depend strongiy on those asmptjons being Correct. fairly simple so there is no longer any excuse tor not exBut few textbooks provide informadon abour how ta d i n g one's data Iar outliers and other ill-mannered derect violations of the normalirg and homogeneity of data. variance assumptions in one's data or what to do about

-

Why Worry? D k c i couqmndeam to Gary Mdlelland. Dcpamncnt ofPsychology, -34% Universiry of Calorado,Boulder, CO 803090345 (e-mil:gary.mdeIlaodQmlorad~~edu).

&¶anypsychologists. qeciaily those far removed from their last smtistio course in graduate school,

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland

Nasty Data – Gary McClelland