|
General Comments concerning inferential statistics in SPSS.
Many analyses are available in the Analyze menu from simple correlations to
multivariate design and more that are available through syntax.
However, one should not let SPSS's options dictate what analyses are
performed. SPSS does not provide much in the way of statistical analysis
post-1975 and so other packages may be necessary to accomplish one's tasks with
more statistical power. But what SPSS may lack in more modern analysis it makes
up for in performing the most rudimentary forms of analyses available with
relative ease (if you prefer clicking 15 things versus typing one line of code
in some packages; I personally find the latter more rewarding).
For example, load the Cars data set which is available with every
installation of SPSS in the Samples directory. Once loaded, one might suspect
there is a relationship between the cars' weights and their mile per gallon
gasoline consumption. To run the linear regression is rather easy, simply
click on Analyze, then Regression, then Linear....
Next, select weight as the independent variable and mpg as the dependent
variable. Then click the OK button.
You should see output similar to what is below.
We would also like to take a look at the
relationship graphically. SPSS has always been fairly weak with regard to
graphical display relative to other programs, with some of the graphs bordering
on laughable (e.g. the default 3d scatterplots). The Graphs menu is easy to use
until you want to tweak and tailor the graph to your own liking, in which case
you're likely in for a headache (and bugs). One doesn't have a whole lot of
control of the initial output nor can one manipulate it very easily. However,
an example is given below of a simple scatterplot from the above examination. A
couple of things stick out. One is that there appears to be a curvilinear
relationship rather than strictly linear one (there are actually subgroups in
this data with linear relationships of varying degrees), and secondly, one case
appears to be an extreme data point (lower left) that will require some action. In this case, the data point is a miskey of some kind (4 cu inch engine?) and
has missing data on several of the other variables and so we would not want to
include it in the analysis.
At any rate, here is how to get a standard scatter plot.
First, go to Graphs in the tool bar, then Legacy Dialogs, then Scatter/Dot...
Next, specify Simple Scatter (default). Then move "Miles per Gallon [mpg]" to
the Y Axis box and move "Vehicle Weight (lbs.) [weight]" to the X Axis box. Then
click the OK button.
You should now see something similar to what is below in the output window;
which is all well and good -- a basic scatter plot showing the relationship
between two variables.
However, it is often desirable to have an actual line of best fit
superimposed on the data, or you may prefer different colors or scale values, or
tick marks, etc. Just about anything can be manipulated in a graphical display
by right-clicking on the graph and entering the chart editor (In Separate
Window).
In the chart editor, just about anything in the graph can be clicked on and
altered. For instance, right click on the data points displayed in the scatter
plot (in the chart editor). When you do, the points should be highlighted in
yellow. You can then select "Add Fit Line at Total".
Once the fit line has been specified, you should see it in the chart editor.
You will also notice, a linear fit line is not the only type of line which can
be specified. Looking at the Properties box (right) shows different types of fit
methods available.
Now, if we simply click (left-click) outside the chart editor, somewhere in the
white-space of the output window; you'll see the line has been applied to the
actual scatter plot.
Other elements of the graph can be changed; color and type of data points, as
well as scale by using the chart editor in similar fashion as what was done
above.
Below, we have the same graph, with changes applied (including a Loess fit
line).
Oneway ANOVA after using recode function to
select only cases with 4, 6, & 8 cylinders. The boxplot below was produced
using the Explore function with the newly recoded filter variable for number of
cylinders.
To conduct the oneway ANOVA, simply go to Analyze, Compare Means, One-Way
ANOVA...
Next, specify your Dependent variable(s) and Factor.
Next, click on the Post Hoc... button to specify which post hoc test(s) and
which equal variances not assumed test(s) you would like. Then click the Continue
button.
Next, click on Options... and select all desired options. Then click the
Continue button, then click the OK button.
The output should look similar to that displayed below.
The Means Plot displayed above was specified when selecting options for the
ANOVA. It is displayed here as a comparison to the boxplot produced above as
part of the Explore function. Clearly, the boxplot is better at conveying the
same information.
|
