This preview shows page 1. Sign up to view the full content.
Unformatted text preview: Which Statistical Procedure is Appropriate?
Variables One categorical variable with 2 levels One categorical variable One numerical variable One numerical variable One numerical variable Major assumptions Test Binomial test SPSS procedure Nonparametric tests Binomial Example research question Comments Tests that ask whether a single variable is distributed as hypothesized None None None None Variable is normally distributed Are 50% of all students Determines if each category are female? occurs as often as expected Are all eye colors are Determines if each category equally common? occurs as often as expected Overly sensitive to sample size; Do IQ scores have a normal probability plot is normal distribution? preferable Do IQ scores have a normal distribution? Do IQ scores have a mean of 100? Plot will approximate a straight line if distribution is normal Determines if population mean is equal to a "test value" specified a priori "One-way" Nonparametric tests chi-square Chi-square KolmogorovNonparametric tests Smirnov 1-Sample K-S test Normal Graph probability Q-Q plot One-sample Compare Means t-test One-Sample T Test Tests that ask whether two variables are related At least 80% of the Two categorical "Two-way" Descriptive Statistics "expected values" are variables chi-square Crosstabs 5 or more Two ordinal variables, or two numerical variables, or one ordinal and one numerical variable Two numerical variables Variables are montonically related Spearman Correlate correlation Bivariate Spearman and Pearson Are eye color and hair correlations are better, if color related? appropriate Does education increase as SES increases? Does IQ increase as height increases? Is the median Compares medians; t-test is crystallized IQ equal to better, if appropriate the median fluid IQ? Is the mean crystallized IQ Compares means equal to the mean fluid IQ? Pearson correlation is better, if appropriate Variables are linearly Pearson Correlate related; residuals are correlation Bivariate normally distributed Tests that ask whether two different variables have the same average Two ordinal variables None "Difference" variable must be normally distributed Sign test Pairedsample t-test Nonparametric Tests 2 Related Samples Compare Means Paired-Samples T Test Two numerical variables Which Statistical Procedure is Appropriate?
Variables Major assumptions Test SPSS procedure Example research question Comments Tests that ask whether two different groups have the same average One ordinal DV and MannNonparametric Tests None one categorical IV Whitney 2 Independent Samples with two levels U test One numerical DV Compare Means DV is normally Independent and one categorical Independent-Samples T distributed t-test IV with two levels Test Tests that ask whether three or more groups have the same average One ordinal DV and one categorical IV None Kruskal- Nonparametric Tests Wallis test K Independent Samples Do males and females Compares medians; t-test is have the same median better, if appropriate IQ? Do males and females have the same mean Compares means IQ? Do all eye colors have Compares medians; ANOVA is the same median IQ? better, if appropriate One numerical DV DV is normally One-way Compare Means Do all eye colors have Compares means and one categorical distributed ANOVA One-Way ANOVA the same mean IQ? IV Tests that ask whether two or more independent variables in a factorial design affect the mean Does eye color vary Two or more Loglinear Loglinear None with gender and hair categorical variables analysis Model selection color? General Linear Model DV is normally Is IQ related to gender, One numerical DV Analysis of Univariate distributed; equal eye color, and the and one or more variance or number of participants interaction between categorical IVs (ANOVA) General Linear Model in each group gender and eye color? Repeated Measures DV is normally distributed; equal One numerical DV, number of participants one numerical IV Analysis of in each group; General Linear Model ("covariate"), and one covariance covariate is linearly Univariate or more categorical (ANCOVA) related to DV; effect of IVs covariate is same for all groups One numerical DV and one or more categorical or numerical IVs IVs are linearly related Multiple Regression to DV; residuals are regression Linear normally distributed Can IQ be predicted Inclusion of covariate "controls" from height, controlling for its effects; hierarchical multiple for SES? regresion is better Is IQ related to height, eye color, and the height x eye color interaction? Which Statistical Procedure is Appropriate?
Variables One numerical DV and one or more categorical or numerical IVs Major assumptions Test SPSS procedure Example research question Comments IVs are linearly related Hierarchical Regression to DV; residuals are multiple Linear normally distributed regression Effect of covariate can vary with Can IQ be predicted group (differs from ANCOVA, from height, controlling which assumes covariate has for SES? same effect in all groups) Tests that ask whether two or more variables represent a smaller number of underlying variables All variables are monotonically related Two or more ordinal How many underlying to each other (if variables, or two or Factor Data reduction Use Spearman correlation matrix factors are there in a 20ordinal), or all more numerical analysis Factor as input if variables are ordinal question instrument? variables are linearly variables related to each other (if numerical) Tests of reliability All variables are How consistent are the Two or more Cronbach's Scale linearly related to each items in a 20-question Measures internal reliability numerical variables alpha Reliability analysis other instrument? ...
View Full Document
This note was uploaded on 04/17/2008 for the course PSC 3000 taught by Professor Sherman during the Spring '04 term at UC Davis.
- Spring '04