At times, SPSS seems to come out with new versions faster than rabbits.
And while sometimes that’s annoying (especially when your organization won’t upgrade their site license, even though all your collaborators’ organizations did), SPSS has added some really great functionality to recent versions.
From a researcher’s point of view, the best one in version 17 was adding Multiple Imputation to the Missing Values Analysis module. It’s an add-on module, so many site licenses don’t contain it, but I encourage you to lobby your site license administrator to ante up for it. (No, I don’t get any kickbacks for mentioning this). It has the best missing data diagnoses of any software I’ve used, but missing multiple imputation (pun intended) was a major disadvantage.
Version 18 had more subtle improvements, like new nonparametric tests.
But IBM is planning some exciting additions to version 19, most notably Generalized Linear Mixed Models (GLMM). This might not sound exciting to you yet, but it is to me.
In case you don’t know, generalized linear models are models in which the error term follows something other than a normal distribution–a Poisson, binomial, or gamma distribution. Common generalized linear models are logistic regression for categorical response variables and Poisson regression for count response variables.
SPSS can already do these.
The Mixed part comes in when you have a nested data structure–children nested within classes, employees nested within organizations, repeated measurements nested within individuals. In other words, all repeated measures designs have a nested data structure.
SPSS can already do these, too, for continuous response variables.
What SPSS currently can’t do is run models in which the response in a model model (often repeated measures) is not normally distributed–binary, ordinal, or count. No matter how many transformations you try, you’ll never get normal residuals from these kinds of response variable.
As of now, the only options are to either run a GEE model (Generalized Estimating Equations) or run a GLMM in another statistical package. They’re available in HLM, SAS, and Stata, among others.
For the most straightforward nested designs, a GEE model works fine. But they’re limited. You can’t, for example, run a random slope model, a three-level model, or a crossed random effects model with a GEE.
Just this past fall, I had a client who needed to run a crossed random effects logistic regression model. I had to run it for her in SAS, which she had never used.
Just to be clear, GLMMs are not easy stuff–they require some heavy math and you do need to know what you’re doing. But when you’re working on a new, heavy statistical procedure that is just beyond your comfort zone (even with guidance), is not the time to have to break into a new, heavy statistical software package.
So let’s hope the new GLMM procedure in SPSS makes in through beta (I’ll be testing it). It’ll make SPSS that much more useful to researchers.
{ 14 comments… read them below or add one }
SPSS 19 disappoints:
http://www.spssforum.com/viewtopic.php?f=5&t=5581
It is all well and good the stuff they keep adding, it is the other things that they take away, basic things like copying and pasting. Have you tried it in SPSS 19? You cannot copy and paste without losing the information at the top. You cannot copy one cell to another. It is mind blowing that with all the other statistical abilities this program seems to have, it cannot do basic stuff, what else have they left out. I know our corporation is not certifying this version of SPSS. Currently use 17 and now having to move to 18.02 couldn’t even get 18 right either. Not holding out a lot of hope for 19.
No, I never did try 19, unfortunately. But I’ve been using 18.0, and I loved it. The improvements they made to the syntax make it MUCH easier to use.
I skipped 17 entirely, and I remember hearing similar things about 16, and people switching back to 15.
It is definitely hard to get excited about improvements if they mess up the things you used to like about a program.
The copying and pasting issue in v19 is incredibly annoying. Not only is the case that indeed, when you paste a series of rows on a cell anywhere other than the top of the column, it incomprehensibly decides to past the same values beginning at the top of the coumn, but when pasting from other sources, say Excel, missing values are replaced by zeroes. Any ideas on how to get IBM to fix this nonsensical behavior?
It is unbelievable that you can’t copy/paste directly from one cell to another. You have to a) ensure Value Labels are turned off, b) Copy/Paste the numeric values into Excel, c) then copy/paste FROM Excel BACK INTO SPSS, d) turn Value Labels back on. I’m sure the devs think they have a valid reason for doing this — but whatever that reason is, this has definitely put a BIG wrench into the works for me.
I confirm the issue about copying and pasting clearing the variables already entered on the first case… (for subject #1, in my database)
How can this be?? Are we talking about a free, open source project on its early stages, or SPSS, the tool I learn back in 1991 as the most reliable way of doing serious statistics work?!
What about IBM?
I am used to see IBM as the last source for reliability in computing…
I think this is such a basic error that it was missed by QA at IBM (who would think that something like this wouldn’t be assured by any trainee at IBM)?
Make no mistake, and no excuse: changing data a person has entered in one part of the database because a person is copying and pasting data at another is PLAIN WRONG! If anybody thinks it has a reason, that person is wrong, and offending the SPSS and IBM brands…
I myself feel my confidence severely damaged, and I am considering continuing my data entry on Excel: I am in the process of coding a database of 300 variables x 800 subjects for my PhD, and came across this issue when I was trying to copy the data that is the same from my subject #84 to subjects 85-121, in order to just change the fields that are different…
And, for my surprise I saw subject #1 variables were set to “0” !!!
What more can I not trust??!
I am very sad at this, for several reasons…
Best regards,
Luís
Please IBM, live up to my expectations…
I haven’t tried the new version, because I can’t even download the new trial version. The IBM website is somewhat confusing in that aspect. I contacted their support. The tech was really nice and tried to go through it, but then realized that he also could not figure out how to download and install the trial!!
Considering what they charge for a license, I certainly will not be moving on to v19 if I can’t even try it out, and have read several disappointing comments.
As an administrator that has to run monthly, quarterly, and annual reports on hundreds/thousands of rotating clients, I was flabbergast when I came across the copy/paste “error”, or so I thought at the time after recently updating to SPSS 19. I searched and even called my IT department for answers until I stumbled across this site to find, to my absolute amazement, that my worst fears were confirmed. They eliminated the copy/paste function, or shall we say, nerfed it to oblivion? What could have prompted this infuriating decision by IBM? I am almost at a loss of words for this extremely poor decision in reducing the copy/paste functionality within SPSS 19, and because my University is requiring a mandatory update to version 19 by October, I’ll essentially have no choice but to either live with this idiotic stripping of a very essential set of functions, or I’ll have to go back to excel. Why IBM…WHY?!?
There is a downloadable “fix kit” 19.0.0.1 on the SPSS website that resolves this issue!
this is ridiculous – in 20 years this software has randomly lost or gained functionality when each version appeared- but this makes it almost unworkable-
@Peter Csizmadia thanks very much for the excel tip- at least i can work- but how anyone thought this up is beyond me
clowns
Does this resolve all copy and paste issues? I not only have the above problems where it changes cells above, but I just atttmpted to paste in some data from row 20 onwards (as do not have data for 1st 20 rows yet) and I highlighted the exact cells in which I needed the data to paste…. it ignored my highlight and pasted it from row 1!!! So pasted the data in completely the wrong place.
This version makes me so angry!!!!!! I also feel really nervous about other problems I might not be aware of and am really nervous about my data such that I check and recheck which is taking so much time.
My company brought seven SPSS statistics 19 licences and we had to revert them all back to 18 because of this issue – what a waste of time and money. Data integrity is more important than any new add-on that comes with 19 or now 20. While I realise that patches have since been released, we cannot afford to jeopardise our work processes by any additional product failures.
I mistakenly thought a company like IBM would have a firm understanding of the importance of rigorous product testing. We blindly purchased every new version of statistics since version 13, that will no longer be the case.
Hi, I’m not with good math knowledge. I’d like to ask a question. Im testing a relationship (IV = a binary variable Risk-taking/risk-averse behavior (measure at time 1); and DV = Change in performance (measure at time 1 and 2); control variables = work hours, work commitment). Can I use spss 19 to see if this relationship is significant and how? My mentor suugested me to run a GLS model in Stata which I’m really not familir with. I understand this place my not be one that I can give this question bu I’d be very muc appreciated if you can give me some hint
Hi David,
Maybe. You could run a GLM in SPSS 19, or a mixed model, which uses REML by default.
GLS is generalized least squares. I believe it is Stata’s default approach to running mixed models. So when your mentor asks for a GLS model, I’m not sure if that means a mixed model, or if there is some reason for preferring GLS specifically.
The GLS or REML refers to the estimation technique. Most of the time, you get nearly identical results using GLS and REML, off maybe by rounding.
I’d have to know more about exactly what your mentor is looking for and the details of the design to be much more helpful. The devil is always in the details!
Karen