SPSS Statistics

 View Only

  • 1.  Shapiro-Wilk en SPSS

    Posted Thu March 13, 2025 11:18 AM

    Hello, I am using SPSS to perform the Shapiro-Wilk test. I input data with a sample size of less than 50 and export the results to PDF using the menu option: Analyze > Descriptive Statistics > Explore. I leave the default settings so that the software exports the results. For example, I input the following 10 data points:

    Copy
    56.0
360.9
391.5
395.8
476.0
477.0
481.0
790.0
835.0
988.0

    SPSS generates a Shapiro-Wilk statistic of 0.925 and a significance level (p-value) of 0.402. However, when I consult the Shapiro-Wilk coefficient table, I get the same statistic (0.925), but the significance level is very different, showing as 0.842 with a 5% significance level.

    My questions are:

    1. Why is this value (p-value) so different?

    2. Is there a way to configure SPSS to use a 5% significance level?

    Thank you very much for your support!



    ------------------------------
    Juan Quik
    ------------------------------


  • 2.  RE: Shapiro-Wilk en SPSS

    Posted Thu March 13, 2025 11:23 AM
    You don't configure SPSS to use a specific sig level.  The test gives you the sig level it finds.

    I suggest, though, that you install the STATS NORMALITY ANALYSIS extension command via the Extensions > Extension Hub menu.  It gives you a number of univariate and multivariate normality tests and diagnostic plots useful in diagnosing nonnormality.  As a general matter, many people prefer the Anderson-Darling test, which is included in NORMALITY ANALYSIS, over Shapiro-Wilk test.


    --



    Original Message


  • 3.  RE: Shapiro-Wilk en SPSS

    Posted Fri March 14, 2025 09:36 AM

    Hello Juan.  You have not said why you want to use a normality test.  Sometimes, people think it is necessary to test for normality of the DV before carrying out a t-test or ANOVA model, etc.  If that is why you are testing for normality, my advice to you would be, don't bother!  I addressed this issue in a short conference presentation several years ago. 

    I hope this helps.



    ------------------------------
    Bruce Weaver
    ------------------------------

    Original Message


  • 4.  RE: Shapiro-Wilk en SPSS

    Posted Mon March 17, 2025 05:10 AM

    Bruce, the title of your presentation suggests that there might be more of them? Is there a number 2 of "silly things we often do"? It would be interesting to read.

    As a small addition, I found these references in David Moore's Basic Practice of Statistics (my third edition is now old, but I expect the 9th edition to cover the same topic) with respect to the robustness of the t procedures:

    • Posten, H.O, "The robustness of the two-sample t-test over the Pearson system", Journal of Statistical Computation and Simulation, 6(1978)
    • Posten, H.O, Yeh, H. and Owen, D.B., "Robustness o the two-sample t-test under violations of the homegeneity assumption", Communications in Statistics, 11(1982).

    It is quite interesting to see that these rather old findings and the claims in your presentation have not become common knowledge yet.



    ------------------------------
    Robert Lundqvist
    ------------------------------

    Original Message


  • 5.  RE: Shapiro-Wilk en SPSS

    Posted Mon March 17, 2025 09:07 AM

    Hello Robert.  There are two other "silly or pointless" presentations:

    • Silly or Pointless Things People Do When Analyzing Data: 2. Using the Wilcoxon-Mann-Whitney Test to Deal with Heterogeneity of Variance
    • Silly or pointless things people do when analyzing data: 3. Transforming variables to make them more "normal" prior to linear regression analysis

    You can find them (and some other presentations) by clicking on the Presentation link on the left side of the window here. 

    Cheers,
    Bruce



    ------------------------------
    Bruce Weaver
    ------------------------------

    Original Message


  • 6.  RE: Shapiro-Wilk en SPSS

    Posted Mon March 17, 2025 04:28 AM
      |   view attached

    Hi Juan

    I tried the Shapiro Wilk test for your dataset with the NORMALITY ANALYSIS extension procedure and getting the same p-value of 0.402. Even you can have a look at the attached screenshot for the p values obtained by other methods. Like Jon Peck mentioned, I too suggest you to use  the NORMALITY ANALYSIS extension procedure for more tests of normality. But note that we need to select at least two variables here but you will obtain the output of both univariate and multivariate case separately. Regarding the comment on normality test as a pre requisite, I always recommend to check normality test before proceeding with t-test or ANOVA test or the similar ones because statistically lot of reasons are there supporting this.

    Thanks



    ------------------------------
    Bindu Krishnan
    Senior Statistician
    IBM SPSS Statistics
    ------------------------------

    Original Message


  • 7.  RE: Shapiro-Wilk en SPSS

    Posted Mon March 17, 2025 09:46 AM

    Hello Bindu.  This sentence in your recent post caught my eye:

    Regarding the comment on normality test as a pre requisite, I always recommend to check normality test before proceeding with t-test or ANOVA test or the similar ones because statistically lot of reasons are there supporting this.

    I gather you are referring to and disagreeing with the tongue-in-cheek presentation in which I argued that it is both silly and pointless to rely on statistical tests of normality to justify use of a t-test or ANOVA etc.  If so, that's fine.  You are certainly free to disagree.  But I am curious about what specifically you disagree with.  For example: 

    1. Do you not agree that statistical tests of normality have too little power when n is small and too much power as n increases?  
    2. Do you not agree that for OLS models (including t-tests & ANOVA), normality of the errors is a sufficient normality condition, but the necessary normality condition is approximate normality of the sampling distributions of the parameter estimates?  (I say approximate normality, because I agree with the things George Box said about straight lines and normal distributions in the real world in his 1976 article--see the excerpt pasted below.)
    3. Can you provide references supporting use of statistical tests of normality as prerequisites to t-tests, ANOVA, etc.?  (I ask because the only ones I remember seeing are in introductory to intermediate level textbooks written by non-statisticians.)
    4. When you do use statistical tests of normality, do you use them on the outcome variable, or on the residuals from the model of interest? 

    Thanks for clarifying.

    Cheers,
    Bruce

    Excerpt from Box (1976)

    In applying mathematics to subjects such as physics or statistics we make tentative assumptions about the real world which we know are false but which we believe may be useful nonetheless. The physicist knows that particles have mass and yet certain results, approximating what really happens, may be derived from the assumption that they do not. Equally, the statistician knows, for example, that in nature there never was a normal distribution, there never was a straight line, yet with normal and linear assumptions, known to be false, he can often derive results which match, to a useful approximation, those found in the real world.

    https://www-sop.inria.fr/members/Ian.Jermyn/philosophy/writings/Boxonmaths.pdf



    ------------------------------
    Bruce Weaver
    ------------------------------

    Original Message


  • 8.  RE: Shapiro-Wilk en SPSS

    Posted Mon March 17, 2025 10:35 AM

    This looks like a formal response regarding a normality test analysis in SPSS or a similar statistical software. Here's a more polished and professional version of your message:

    Subject: Normality Analysis using Shapiro-Wilk Test

    Hi Juan,

    I performed the Shapiro-Wilk test for your dataset using the NORMALITY ANALYSIS extension procedure and obtained the same p-value of 0.402. You can also refer to the attached screenshot, which shows p-values from other normality test methods.

    As Jon Peck mentioned, I also recommend using the NORMALITY ANALYSIS extension procedure, as it provides multiple tests for normality. However, please note that at least two variables must be selected for this procedure, but the output will still display results for both univariate and multivariate cases separately.

    Regarding the importance of normality tests, I strongly recommend conducting a normality check before proceeding with statistical tests like t-tests, ANOVA, or similar procedures. There are several statistical reasons supporting this best practice.

    Let me know if you have any further questions.

    Thanks,
    jerry_Tomson



    ------------------------------
    jerry Tomson
    ------------------------------

    Original Message