General Statements on Code

Entering data:  In dealing with entry, I prefer spreadsheets but do know that they can cause problems being read into other programs.  Also, the advantage of doing it in syntax is that it won't matter what version someone is using now or in the near future, as the program creating the dataset will still produce it.  One last note, for those doing online surveys or dealing with archival data, you very likely will be dealing a text document as data is saved in that format because all programs can read it.

Advantage: SPSS spreadsheet (pre-16)

Importing data from other stat programs:  All are typically good at reading other program files, and assuming the original creator of the data didn't cut corners, it can easily be read in via menus.  You almost always have to make the choice for labeled data on whether you want the labels or the numbers (can't have both; e.g. gender coded as 0 and 1)

Advantage: None

Manipulating Data: Depends on the program really, but as I've gotten used to it R is turning out to be much easier and faster in that department than SPSS (menus or otherwise).  Still looking into SAS. 

Sort: Note, you should ask yourself why you need to sort. If you want to by ID I can understand, but most do it because what they really want to do is subset the data and they don't know how to in an efficient way, so they sort and then start cutting up the data . It's also much easier to go to a specific case or variable by other means (e.g. ctrl+F in a spreadsheet) than sorting and scroll-hunting.

Merge: Merging is trivially simple when datasets match up, but can be extraordinarily difficult when they don't for some programs.  I've seen people use different names for the same variables, use different ID systems etc. and it's just laziness, and because they cut corners or don't take the minimal effort needed they end up adding hours to this process.

Split: Not a big deal for most packages.  SPSS's default approach is terrible via code and would be much more easily done via menus.  However if you use the 'temporary' command it's easier than menus and is to be preferred also because it will 'turn itself off' after analysis of choice is run.

Subsetting data:  I find R much easier than SPSS as it is as simple as newdata=olddata[rowsyouwant, columnsyouwant], and it can't really be done any easier than that.  Selecting data based on values is also easy, but I find all three packages can do the typical stuff like that with a DO IF type of statement.

Computing new variables:  depends on the situation but they can all do it pretty easily, but R has a built in a suite of operators for calculations on matrices. Because of this it is extremely easy to do any sort of computation, and there often is nothing special needed, i.e. you don't need a special 'compute' command to create new variables, and in general its capabilities are more advanced than the other two programs.

Apply: the apply function in R is very powerful, especially for computing functions over levels of a factor or across variables.

 

Frequencies:  Frequencies are handled with about the same amount of ease regardless of program, however with I don't think it gets any easier than the table function in R, whose output can then be easily used for graphing or other purposes.  SPSS also by default assumes charts will be of interest.

Advantage: None

Central Tendency and Variability:  while getting simple means and standard deviations is easy for any program, getting modern statistics that one can actually use is not to be had in SPSS.  Both SAS and R provide easy means for obtaining robust measures to one's own specifications.  R is unparalleled in robust offerings in general however, and getting geometric or harmonic means is also easy.

Advantage: R

Summary stats: SPSS really needs to get rid of having both Descriptives and Explore menus, there is no reason for both.  Explore is ugly and for some reason is called 'Examine' in syntax, but much more flexible and has more options.  However R and SAS can get the same sorts of things with minimal code, and with R, the output is automatically accessible by other functions for further manipulation.

Advantage: R, SAS

t-tests: SAS I don't have up yet but have glanced at some. All seem to be quite capable here providing CIs , automatic unequal variance solutions for the independent samples case etc.  SPSS fails however at allowing a specification for one-sided tests or non-nil hypotheses, SAS also seems to fail in these respects.  This doesn't seem like much but in terms of the issues regarding NHST, we know better.  One thing I like about SPSS is the ability to do multiple t-tests very easily.

Advantage: R

Correlation: They all give you simple correlations, partial correlations, rank correlations etc.  SPSS has no native ability to do polychoric or tetrachoric correlation, and you can forget robust approaches also (outside of the rank ones).  SAS has polychoric correlation as a default descriptive (via PROC FREQ) and robust capabilities via IML.  R has all the correlations I've come across/wanted, and usually very easily obtained.

Advantage: R > SAS >>> SPSS

Regression:  Let me put it this way- Teaching you regression with SPSS menus is almost unethical in my opinion.  The only reason to do it with SPSS is because you've been asked to by someone else, and I would never recommend using SPSS regression for your own research.  The very fact that it doesn't even have a way to easily test the assumptions of basic linear regression analysis tells you all you need to know, but also its model graphics are something you have to do from scratch as a separate enterprise, and its available regression techniques for analysis appear to be about 30+ years behind.  For a basic linear model, it would take you longer to do everything SPSS offers (which isn't much) than it would to do it right with a modern approach in R.  A list of some of the stuff you can do in R.  Here for SAS.

Advantage: anything but SPSS.

Power and Effect Size:  Statistical Power:  Unless you want to shell out $500 for SamplePower, for some reason SPSS has not figured out to provide anything but provide 'observed' power in its basic package, which is a near useless statistic for most purposes, though it is often used as an excuse for missed effects by poor researchers (see Hoenig, John M. and Heisey, Dennis M. (2001) The abuse of power: The pervasive fallacy of power calculations for data analysis.  The American Statistician, 55, 19-24 ).  Furthermore SamplePower doesn't do any more than the SAS does through PROC POWER, which comes at no additional cost.  R's offerings are also extensive, and very easy for typical analyses via the MBESS package.  I still like Gpower, and version 3 is about as easy as it gets.

Advantage: anything but SPSS.

Effect size: There are a lot of effect sizes out there, so it's difficult to make a general statement about package offerings.  I'll speak a little about commonly used ones.  SPSS offers R-square and adjusted R-square for regression, but while it offers the subsequent Eta-squared for ANOVA, it does not provide the bias-adjusted Omega-squared.  It does not offer Cohen's d for simple comparisons, and I am unaware of any offering for interval estimates for any effect size.  SAS, based on my limited knowledge, does not appear to be much better.  R however, can calculate effect sizes and interval estimates of them based on traditional approaches and bootstrapped.  The MBESS library is a good place to start, as power and effect size is its primary function.

Advantage: R