Hi Jeffrey, thanks for the compliment! If you like our work, perhaps consider taking our full course at bit.ly/spss-beginners-course It includes way more content than our KZbin channel and we'll offer some personal assistance as well. Happy analyzing!

Thanks for the compliment! If you like our work, perhaps consider taking our full course at bit.ly/spss-beginners-course It includes way more content than our KZbin channel and we'll offer some personal assistance as well. Happy analyzing!

Yes, I do. However, I don't cover anything Python. If you purely want to predict stuff, Python may be better. But SPSS is more useful if you want to explain (rather than predict) what's going on in your data. Use a hammer for nails, a screw driver for screws ;-)

You can enable this on KZbin and make it play slower (audio and video). Click Settings> Playback Speed and Select any number below Normal for example 0.75

Thanks for the compliments! If you'd like to see more videos like these, perhaps consider taking our full course at bit.ly/spss-beginners-course Happy analyzing!!

I hate to say this but outliers are truly one of the "grey areas" in data analysis. Even highly experienced analysts often disagree on what (not) to exclude. I think the best ways to deal with this are -use your common sense and -be very open and explicit on what you decided and why. Hope that (somewhat) helps! Ruben SPSS tutorials

@@SPSS-tutorials Dank je wel... by the way, this is a good tutorial video indeed. 🤪🤪🤪

@@kwoncey4581 Thanks for the compliment, happy to hear you like it! P.s. if you like this one, you may want to check out our full SPSS course at bit.ly/spss-beginners-course

The SPSS syntax introduction you mean? The video version is at kzbin.info/www/bejne/iYvUfn6QgJ2Nmac and the written version is at www.spss-tutorials.com/spss-syntax/ Hope that helps!

Hi Hassan! There's no real "right or wrong" here but IMHO, a single analysis with all predictors is preferable over separate analyses because you want the predictors to compete with each other. But if this were my project, I'd first see if I can reduce the number of IV's with PCA ("factor analysis"). Note that you'll usually want at least some 10-15 independent observations (participants) per predictor. Now, if you dummify some of your nominal predictors, you may have more than 40 for 600 observations. Not ideal. Last but not least: perhaps consider doing this hierarchically: enter all subjective IV's in step 1 and the others in step 2 and see if the increase in r-square is statistically significant. Then enter the 2 sets of predictors in the opposite order too. Hope that helps! SPSS tutorials

Hi Vitalis! I had a very close look at this tool and -sadly- SPSS no longer supports it. However, note that recent SPSS versions include CI's for Pearson correlations under Analyze - Correlate - Bivariate (fourth button in right top corner). This requires raw data but if you don't have these, you can use the second sheet in www.spss-tutorials.com/downloads/pearson-correlation.xlsx Just replace r and n and the formula will recalculate the CI for you. Hope that helps! Ruben SPSS tutorials

Yes and no. Strictly, Likert scales are ordinal variables. This implies you can neither compute sums/means over them, nor use them as "normal" (that is, quantitative -note this is slightly different from "continuous") predictors in linear regression. Less strictly, however, many analysts treat Likert scales as if they were quantitative variables under the "assumption of equal intervals". So what you're doing is fairly common practice (especially in non academic research) but strictly not entirely correct. In short, there's no clear yes/no answer to your question. Hope it helps anyway.

Hi Georgios, thanks for the compliment! Your comment makes perfect sense: collinearity is not caused by high correlations among the DV and the IV's. So the short answer is: no, that's not always problematic. However, in the social sciences, super high correlations (say |r| > .80 / .90) are oftentimes suspicious. If we track down where they came from, it's often that 2 variables are basically the same thing such as some sum score computed in 2 different ways (with/without missing values for instance). And that obviously is problematic... So if you see super high correlations, please do a careful data check to see if these truly reflect some linear relation. Hope that helps!

Hi Sonia! This video is not about stepwise regression or any other form of hierarchical regression. For reporting such analyses, there's tons of different APA table formats (most of which are not very useful in my opinion). A useful book for these tables is www.amazon.com/Presenting-Your-Findings-Practical-Creating/dp/143380705X So there's no short or simple answer but we sometimes report -variables entered in each step -R-square change and the associated F- and p-values for (sets of) predictors. Hope that helps! SPSS tutorials

@@SPSS-tutorials thank you so much sir. Your answer is helpful.

@@soniamanhas4970At your service, Sonia! P.s. do check out the book I linked to, it's super helpful.

Hi Jon, thanks for the suggestions! I'm very interested in your SPSSINC MODIFY TABLES suggestion. I created a heatmap from a correlation matrix a long time ago (which I found very interesting) but doing so was hard. So an easier solution for this is super welcome. The matrix scatter is also a good idea for a small number of variables. The tool I used can also add a CURVEFIT for each variable pair and I think that should be a standard routine (but it was too much detail for this simple video). I hadn't even thought about multiple testing. Is there any easy was to apply Bonferroni(Holm) to Pearson/Spearman correlations in SPSS? I guess this is easier for the (more recently introduced) CI's as you can choose any confidence level here such as (1 - .05/6) for 4 variables. But that's just Bonferroni... P.s. I think it's a pity that the latter procedure results in a separate table with a somewhat inconvenient table structure. I'd much rather prefer a standard (square) correlation matrix with CI's in square brackets if requested. IMHO, there's a lot of room for improvement here.

SPSSINC MODIFY TABLES (Utilities > Modify Table Appearance) now has the ability to produce a heatmap effect for any table. Here's an example for a correlation table. It excludes coloring the diagonal. It's a little bit complicated, but the dialog box can generate all this. SPSSINC MODIFY TABLES subtype='Correlations' SELECT="Pearson Correlation" DIMENSION=ROWS LEVEL = -1 /STYLES APPLYTO="x < 1" HMAUTOCOLOR=YES USEABS=YES HMSCALE=LINEAR. There is a lot of control over the color range and how it is spread, but HMAUTOCOLOR does a pretty good job without using those features. That dialog box now allows the user to use a color picker or enter rgb components for colors, but unfortunately, the color picker control doesn't remember state on recall due to a bug in the custom dialog builder. If you paste the syntax, though, it pastes the RGB components, so those can be reused without relying on the color picker control. For the CI issue, it's hard to jam all that information into a single correlation matrix display, but the STATS CORRELATIONS extension command does give you a way to get a reasonable display with a flexible combination of variables. image.png If you want to apply a multiple testing correction, STATS PADJUST can do that. It offers six different types of correction.. Bonferroni might be too conservative for a correlation matrix with more than a few variables, but Benjamini-Hochberg would usually be a good choice. There are two versions of matrix scatterplots available - the Chartbuilder GPL (same as the legacy dialog) and Graphboard Template Chooser, which also puts histograms in the diagonal elements. The Chartbuilder matrix scatterplot can apply a linear fit to each box.@@SPSS-tutorials

@@JonPeck100Hi Jon! I had another look into STATS CORRELATIONS. First off, I think it's a pity that it requires a DATASET NAME. I never use (or recommend) using this. It simply gets in my way without being helpful in any way. Also, the warning may deter SPSS beginners from proceeding with STATS CORRELATIONS if they're unaware that DATASET NAME ... will solve this issue. Why don't you have Python apply and remove a (random) DATASET NAME if not applied by the user? Also, I'm not a big fan of the resulting table layout and the inability to choose which statistics I want to see. Why not have users select r / N / p / CI as they see fit? Finally, as a more general point, I think many extensions are quite inconsistent regarding their naming/titles. I only learned that STATS CORRELATIONS is the same extension as "Bivariate with..." because I stumbled upon some .pdf document with a couple of screenshots. Note that these 2 names literally don't have a single word in common.

Thanks for the compliment. We really appreciate it.

Hi Paul, at your service and thanks for the compliments! All videos thus far were recorded using SPSS version 28.

Hi Jon! I tried the variables toolbar tool but it inserted the variable names separated by spaces. Also, why is something like COMPUTE M01 = MEAN(V01 V03). ambiguous and DESCRIPTIVES V01 V03. not? Am I overlooking something here?

Yes, it only does spaces, as I said, but it is still easier than copying variable names out of Variable View. As for the functions, ALL functions use comma-separated arguments. It would be inconsistent and confusing to have some allowing blanks and some requiring commas. And, remember, the function arguments could themselves be complex expressions.

@@JonPeck100 I find it inconsistent that SPSS procedures allow spaces between variable names but functions don't. Why can't I ALWAYS separate variable names by spaces, both for functions and procedures? Could you give an example of an expression that would be rendered ambiguous by allowing spaces rather than commas? I think the rules for variable names always preclude this -but please show me wrong on this one. Also, I think long variable names are better avoided at all times. Since we've variable labels that accurately describe the exact meaning, long names only render syntax unnecessarily long and unreadable. I oftentimes use something like RENAME VARIABLES (ALL = V000 TO V113). and I never regretted it ;-)

Of the programming languages I am familiar with, they all require some sort of separator for function arguments: Fortran, C, C++, Java, Python, R, Basic, SAS, and others. That is good for readability and makes more accurate error messages possible. If blanks were legal as a separator, argument expressions would not be able to contain spaces. As for variable names, the world rejoiced when SPSS went from 8 bytes to 63 length, but I agree that really long names are clumsy. Name should be long enough to be descriptive for readability, and with the current limit, the user can just forget about a length constraint.

@@JonPeck100 "If blanks were legal as a separator, argument expressions would not be able to contain spaces." Why not? Variable naming conventions always distinguish variable names from numbers, operators and functions, right? So even something like COMPUTE M01 = MEAN(2 * V10 TRUNC ( V20 ) ). is unambiguous insofar as I can see. And I can't think of any example where this would cause any problem... But I agree that this particular example doesn't look nice...