Link

Example data

The image shows the top part of the example data which you can download here

two_way_data

The 4 columns are as follows:

  • factor_1: one of HF or NF
  • factor_2: one of LV_Mid or RV
  • prep: a string that is unique to each row
  • tag: a string that defines the person the data that was obtained from
    • For example, the top 2 rows have the same tag “5CCF6”.
      • This means we have 2 values for HF, LV_Mid from person 5CCF6.this factor is used to define “repeated measures”.
    • If you look further down the data file
      • you can see that we also have 3 values from this person from the RV
  • pCa50: the data we are testing

Two-way model in SAS

Code

The SAS code to run a 2-way linear mixed model is as follows (adjust your paths as required).

/* SAS template for a 2 way analysis with a grouping variable */

proc import out = work.all_data
	datafile = "C:\ken\GitHub\CampbellMuscleLab\howtos\howtos_linear_mixed_models\docs\pages\MATLAB\two_way_model\data\two_way_data.xlsx"
	dbms = xlsx replace;
	sheet = "Sheet1";
	getnames=yes;
run;

ods html file="C:\ken\GitHub\CampbellMuscleLab\howtos\howtos_linear_mixed_models\docs\pages\MATLAB\two_way_model\sas_results\sas_results.html";
ods listing close;

proc print data=all_data;
	title1 'All data';
run;

proc glimmix data=all_data;
	class factor_1 factor_2 tag;
	model pCa50 = factor_1 factor_2 factor_1*factor_2 /ddfm=satterthwaite;
	random tag;
	lsmeans factor_1 factor_2 factor_1*factor_2 /slice = factor_1 slice = factor_2 slicediff=(factor_1 factor_2) pdiff adjust=tukey;
run;

ods listing;
ods html close;

Results

When run in the SAS environment, this produces an output file in html format

Interpretation

The top section of the results file shows the input data.

SAS_input_data

The next section that is critically important is section showing the main effects and their interaction.

SAS_main_effect

Here, the

  • p value for factor_1 is 0.0055
    • This indicates that the HF and NF groups are statistically different
  • p value for factor_2 is 0.919
    • This indicates that there is no statistical difference between LV_Mid and RV
  • p value for factor_1 * factor_2 (interaction) is 0.029
    • This indicates that the effect of factor_1 depends on the value of factor_2

The last section shows the post-hoc tests.

SAS_post_hoc_tests

These show:

  • in HF samples, LV_Mid and RV are statistically different (p=0.0394)
  • in RV samples, HF and NF are statistically different (p=0.0002)