Need some assistance with SPSS. At this time, I'm a novice on the SPSS software package.
Obviously, there's tons of information on Google and YouTube. At the same time, having a specific question, I'd like to reach out to the EE network and see if I can get some general pointers:
Quick Background:
- As part of some research, I have several independent variables (across different categories) and several dependent variables.
- To determine correlation (in SPSS), I believe one of the viable options is to use "One-Way ANOVA"...
- Thus, for testing purposes, I've selected one of the independent variables and a dependent variable.
- When executing the One-Way ANOVA, I don't see any "Multiple R", "R Square", "Adjusted R Square", or "P values", etc. (as seen when computing in Excel).
My question:
Whether a one-to-one (1 independent variable vs. 1 dependent variable) OR many-to-one (several IVs vs. 1 DV), what are some viable statistics/test that I should run in SPSS?
Again, just looking for some general guidelines... SPSS is extremely powerful and I probably won't use more than 10% of its capabilities. Right now, it's just a bit overwhelming when learning this tool.
Hi EEH,
I'm a huge fan of EE, but in this particular case, I suggest going to some experts that specialize in statistics. I previously worked for a company called Straight Line Performance Solutions (there's no conflict of interest, since I no longer work for them and have no financial interest/involvement whatsoever with the company). They are heavy-duty experts in statistics and offer a free service called Ask a Statistician. I have first-hand experience seeing their responses to questions asked at the Ask a Statistician page and I can say that they are excellent...and in most cases, very fast. Regards, Joe
First of all, there is an SPSS section of Experts Exchange - if you want SPSS-specific help, you should post there.
Second, the problem you are having is not with SPSS, but rather with your understanding of statistics. One-Way ANOVA (SPSS or otherwise) is a test that partials total variance in a DV into between-group (IV) and within-group (error) variance. The only effect size estimate it can really produce is eta-squared, which is related to R^2 (multiple R^2), but only in certain cases.
You have mixed up a lot of different analyses with different purposes, all of which are run in different places in SPSS. I will go through what I am interpreting as the issues you are having one by one:
1) "To determine correlation (in SPSS), I believe one of the viable options is to use "One-Way ANOVA"
This is incorrect. A 1-Way ANOVA and correlation are not the same. If you want simple bivariate correlations, you should open Analyze > Correlate > Bivariate and select the pair of variables you are interested in correlating. That prompt is capable of reporting most types of correlation you probably want, whether it is Pearson's, Spearman's, etc.
2) "When executing the One-Way ANOVA, I don't see any "Multiple R", "R Square", "Adjusted R Square", or "P values",
The assumptions of One-way ANOVA do not include linearity, so R and related statistics are not reported. Excel just assumes you know what you're doing and will only interpret the statistics that you should interpret (which is why it's dangerous to use Excel for statistical analyses if you're not well-versed in statistical theory). r-squared is literally that - r multiplied by r. To get an r, you want to calculate a Pearson's correlation, which you can do using the technique described above. You can then multiply it by itself with a calculator to get r-squared.
You also mentioned Multiple R, which is based on a different family of analyses, called linear regression (little r and big R are identical if you have only one IV and one DV, but little r can't be calculated with multiple IVs). Multiple R requires multiple independent variables. If you actually do want multiple IVs predicting a single DV, you can do so in SPSS within Analyze > Regression > Linear. It will report R, R-squared, and p-values associated with tests of the intercept and the slopes of each predictor against zero, depending on which you want, along with a multiple-R indicating total variance shared between outcome and the full predictor set.
3) "Whether a one-to-one (1 independent variable vs. 1 dependent variable) OR many-to-one (several IVs vs. 1 DV), what are some viable statistics/test that I should run in SPSS?"
It depends on what you want to know, and the analyses I will mention below assume you have interval or ratio level scale of measurement on both of your variables. If you don't, you'll need different tests. Given that... if you just want to correlate two variables, a simple Pearson's correlation is probably what you want (use the technique described above). If you want to predict one DV from several IVs (again, all must be interval+), multiple linear regression would be a reasonable analysis (also follow approach above), and you might want to interpret slopes or overall model fit, depending upon your research question.
Also note that if you have group membership data in your IVs, you probably want n-way ANOVA, but it's not going to report R-family statistics, because they are not appropriate for ANOVA.
0
ExpExchHelpAuthor Commented:
Joe -- thank you the pointer to "Straight Line Performance Solutions"... I definitely check it out.
*****
richdiesal -- although I've taken stats classes, I certainly don't consider myself a SME on it. So, I do thank you for providing some general background as to what other/better methods should be explored. Btw, when searching on groups I entered "SPSS"... but it didn't show up. Is it list under its full name (vs. acronym)?
Again, thank you both for your feedback... it's very much appreciated.
0
Modern healthcare requires a modern cloud. View this brief video to understand how the Concerto Cloud for Healthcare can help your organization.
Rich,
I didn't know that there's an SPSS section of EE...thanks for that piece of info! I also didn't know that there are experts on statistics like you...I had the sense that the experts are all computer scientists. :) Thanks for correcting my misimpression. Regards, Joe
Per your recommendation, I've used the bi-variate (Pearson's method) to compare one of the IV in respect to one of the DVs.
My sample data is rather small (i.e., 9) at this time. Eventually, I anticipate to have several hundred data point. So, current data is only used to familiarize myself with the various methods/processes in SPSS.
Per attached JPG, would you mind helping me out to interpret some of the listed information?
Thank you in advance,
EEH
P.S. I didn't expect the bi-variate method has one a single dialog box for the variables. How does SPSS know which is the IV vs. the DV? Pearsons.jpg
0
Featured Post
Modern healthcare requires a modern cloud. View this brief video to understand how the Concerto Cloud for Healthcare can help your organization.
I'm a huge fan of EE, but in this particular case, I suggest going to some experts that specialize in statistics. I previously worked for a company called Straight Line Performance Solutions (there's no conflict of interest, since I no longer work for them and have no financial interest/involvement whatsoever with the company). They are heavy-duty experts in statistics and offer a free service called Ask a Statistician. I have first-hand experience seeing their responses to questions asked at the Ask a Statistician page and I can say that they are excellent...and in most cases, very fast. Regards, Joe