The first version of SPSS was published 59 years ago. The first version of JASP was published three years ago. How does an expensive IBM-developed program compare to the free and open source program which was until recently just developed by a handful of scientists?
My approach for this review is as a university lecturer in psychology. Ease of learning and use matters since it allows more mental resources for statistical understanding. Also, consistency with the publishing guidelines of the American Psychological Association (APA) matters since that’s how students are supposed to write their reports and papers.
How JASP is better than SPSS
SPSS is APA-incompatible
SPSS cannot compute Cohen’s d. Furthermore, SPSS cannot compute confidence intervals on Cohen’s d, Pearson’s r, Spearman’s rho, Mann-Whitney U, Wilcoxon’s matched pairs, Wicoxon’s signed-rank test, Binomial proportions nor show confidence intervals on ANOVA interaction plots. JASP does this in a click of a checkbox. Reporting effect sizes and intervals is not just a cornerstone of APA publishing guidelines; it’s also important for the “new statistics” movement. So don’t use SPSS if you want to publish.
One small click in JASP; a major improvement over SPSS.
Installing JASP is easy and reliable
SPSS requires a campus license which needs to be activated and updated ever so often. In addition, we usually see 3-5 students every year who simply cannot get SPSS working. Students messing up licensing and having install problems cause a lot of wasted time. JASP is free, thus avoiding license problems. I have yet to experience a single student who cannot get JASP running.
Testing assumptions in SPSS is ridiculous
Compare the procedure for testing the normality assumption in a paired samples t-test in JASP and SPSS:
JASP: click “Normality” under the aptly named section “Assumption checks.”
Let’s count the number of clicks to test normality in JASP: ooone… oh, it’s done already!
SPSS: Realize that a paired-samples t-test corresponds to a one-sample t-test of the pairwise differences. Then compute that difference using Data → Compute variable… → diff = var2 – var1. Then head to Analyze → Descriptives → Explore → Plots → Normality plots with test and run the analysis on the newly computed “diff” column. Then scroll down through several tables and see if you can find the test of normality. Seriously, why is the Shapiro-Wilk test hidden under “Descriptive statistics” (it’s inferential!) and “Plots” (it’s a table!)? Try motivating students to test assumptions using SPSS. It’s not a pleasant experience.
SPSS’ output for normailty tests. It’s so long that I had to split it into three columns and scale it down to fit it into this blog post. There sure is a lot of normality checking going on (and then some), but how much of it is relevant for APA-reporting?
SPSS output is bloated and difficult to convert to APA-style
I usually tell my students that the SPSS pedagogy is the following: We will present the results using the worst possible defaults so that users are forced to deeply understand and tweak everything to not make a fool of themselves. It’s great for spotting students who don’t have a firm understanding of statistics and who uncritically copy-paste the output into their reports! But it is cumbersome to teach, making the students waste attention on figuring out where to find the results rather than understanding the results. And really, it’s not the job of educational software to set up honey traps.
- JASP tables APA-formatted by default, containing just the data necessary for an APA-style report. You usually have to add confidence intervals with a click of a button, but that is tolerable. SPSS tables are horribly formatted containing loads of unnecessary information as was clear from the example of normality-test above. There are too many examples to do this point justice, but one favorite of mine is that SPSS repeatedly adds rows and columns to descriptive tables to tell you that 100% of the observations amount to 100% of the observations.
- JASP provides APA figures, i.e., gray-scale figures and sensible axis ticks. SPSS Figures have twelve decimals on all x-axes tics, even for integers (1.000000000000, 2.000000000000, etc.). It happily adds means and standard deviations (with 12 decimals) to histograms of non-normal data, as shown in the figure below. Also, the SPSS color palette’s first three colors prints to the same shade of gray. At our university, most students hand in their reports in gray scale, rendering it near impossible to interpret the plots.
JASP (left) vs. SPSS (right) histograms. Whoa, that’s a lot of decimals, SPSS!
- JASP names statistics by their name (t, W, U, etc.) as APA-style reporting requires. In many cases, SPSS uninformatively call them “statistic,” leaving the students to guess. See e.g. the Shapiro Wilk’s W above.
- JASP has no “variable view” with loads of strange setting such as column width. There is simply a single data view where you set the variable settings in the column headers. Complexity reduced by 50%!
- To conduct an analysis, you don’t have to go through the “Analyze” menu first. The analyses are available straight from the ribbon, skipping one or two steps compared to SPSS.
- Redoing an analysis in SPSS requires either going through all menus again or copying the syntax, pasting to the syntax editor and pressing “run.” Then delete the old unwanted output. To edit a particular analysis in JASP, just click the corresponding output and edit the settings. It is very intuitive to click the analysis you want to edit rather than taking detours through menus. Furthermore, the results will update immediately, reducing the overhead of trial-and-error.
- The latter point above also increases the efficiency of supervision. With SPSS, students often send the results (.spv file), and you sometimes have to look very carefully to figure out what the students actually did or where an error occurred. JASP saves the full data and analysis to the .jasp file, so everything is reproducible and interactive at the supervisor’s end. See something wrong? Edit a few settings, see that it works, annotate the output with your comments, and send it back to the students.
The JASP interface is simpler
It requires substantially fewer clicks to do anything in JASP compared to SPSS. This is important in teaching because the probability that a student gets lost in a demo increases exponentially with the number of steps of the procedure. Notably:
SPSS cannot do meta-analysis. This is Cohen’s d and assumption checks once again: SPSS lacks some basic and often used functions recommended by APA and by pretty much everyone. The upcoming JASP 0.8.4 will feature meta-analysis at a level on par with Cochrane’s widely used RevMan. This will save you the hassle of using a different piece of software for a particular analysis. Furthermore, a look at the JASP source code reveals that more advanced features are in the making, such as meta-regression, trim-and-fill analyses, etc.
Until this point, I have kept my true Bayesian identity in hiding, but the fact that JASP provides Bayesian equivalents to many of the frequentist analyses constitutes a major improvement in science and the teaching of statistical inference. Once you’ve tried explaining research results from a particular study using a confidence interval as compared to a credible interval without the frequent misunderstandings of the former, you know what I mean.
JASP has included ways to set informative priors, but there is room for improvements in the Bayesian analyses, in particular for those more concerned with posterior predictive checks than Bayes Factors as an assessment of model adequacy. Also, posterior distributions of parameter values in regression are not provided yet. Given the current speed of JASP development, I have no doubt that this will come very soon.
JASP is actively developed and responsive to requests
Given the above, SPSS development seems to have halted to a crawl relatively speaking. If SPSS cared for its user base, Cohen’s d would have been included 15 years ago. JASP, on the other hand, is developing at lightning speed and due to its open-source nature, this pace will accelerate once researchers across the world join in on the development. I have presented several ideas to the JASP team on GitHub and many of those were implemented in JASP within 1-2 months.
Where SPSS is (currently) better than JASP
Though SPSS has horrible plotting defaults, the plot options and tweaking options are comprehensive and you can, for the most part, force something nice out of it. JASP only provide a few (nice!) non-editable default plots but nothing else. For example, it cannot make scatterplots. The JASP team is working on a comprehensive solution but have not indicated what it is and when it will arrive.
JASP relies on your spreadsheet editor to do all data editing. While this is a clever solution in many cases (e.g. changing values, computing variables, etc.), it is probably too involved for teaching purposes when converting from wide to long format or from long to wide format. Personally, I hope that most analyses will (also) work on long format, including RM-ANOVA, but probably with an exception for multivariate analyses. JASP has given this issue high priority, but no steps towards a solution have emerged yet.
Doing analyses on subgroups of data can be useful but only SPSS supports this so far, and it would be cumbersome to delete/restore data in Excel, e.g. selecting only Hispanic males.
Install JASP now and use it from this moment and on. Just do it. Keep SPSS hanging around for plotting and perhaps for major restructuring of data for a year or two.
Like this post?
Subscribe to our newsletter to receive regular updates about JASP including our latest blog posts, JASP articles, example analyses, new features, interviews with team members, and more! You can unsubscribe at any time.