-matrix2stata-
Description
The command -matrix2stata- converts roweq, rownames and content from a matrix into variables. This makes it eg possible to do graphs of the content of one or more matrices. The difference from -svmat- is that it adds roweq and rownames as variables for grouping.
The matrices might have been generated from -sumat-.
Installation
To install use the command: ssc install matrixtools
Demonstration
The dataset
The dataset in this example is auto with a value label:
sysuse auto, clear
label define rep78 1 "1 repair" 2 "2 repairs" 3 "3 repairs" 4 "4 repairs" 5 "5 repairs"
label values rep78 rep78
The command -sumat- is used to generate a matrix with some content to plot:
sumat price if foreign == "Foreign":origin, statistics(mean ci) rowby(rep78) roweq(Foreign) full
matrix out = r(sumat)
sumat price if foreign == "Domestic":origin, statistics(mean ci) rowby(rep78) roweq(Domestic) full
matrix out = out \ r(sumat)
And the matrix looks like
matprint out
--------------------------------------------------------------------
mean ci95% lb ci95% ub
--------------------------------------------------------------------
Foreign Repair record 1978(1 repair)
Repair record 1978(2 repairs)
Repair record 1978(3 repairs) 4828.67 3373.89 6283.45
Repair record 1978(4 repairs) 6261.44 5022.69 7500.20
Repair record 1978(5 repairs) 6292.67 4485.82 8099.51
Domestic Repair record 1978(1 repair) 4564.50 3840.29 5288.71
Repair record 1978(2 repairs) 5967.63 3487.30 8447.95
Repair record 1978(3 repairs) 6607.07 5226.06 7988.09
Repair record 1978(4 repairs) 5881.56 4841.46 6921.66
Repair record 1978(5 repairs) 4204.50 3772.33 4636.67
--------------------------------------------------------------------
Now -matrix2stata- is used to move the content of the matrix out to the dataset. The option clear clears the dataset before inserting the matrix.
matrix2stata out, clear
The command -matrix2stata- returns the following stored results:
display "`:r(macros)'"
variable_names
display "`r(variable_names)'"
out_eq out_names out_mean out_ci95__lb out_ci95__ub
The current dataset now looks like:
list
+------------------------------------------------------------------------------+
| out_eq out_names out_mean out_ci~lb out_ci~ub |
|------------------------------------------------------------------------------|
1. | Foreign Repair record 1978(1 repair) . . . |
2. | Foreign Repair record 1978(2 repairs) . . . |
3. | Foreign Repair record 1978(3 repairs) 4828.6667 3373.8855 6283.4478 |
4. | Foreign Repair record 1978(4 repairs) 6261.4444 5022.6874 7500.2015 |
5. | Foreign Repair record 1978(5 repairs) 6292.6667 4485.8221 8099.5113 |
|------------------------------------------------------------------------------|
6. | Domestic Repair record 1978(1 repair) 4564.5 3840.2933 5288.7067 |
7. | Domestic Repair record 1978(2 repairs) 5967.625 3487.3029 8447.9471 |
8. | Domestic Repair record 1978(3 repairs) 6607.0741 5226.0617 7988.0865 |
9. | Domestic Repair record 1978(4 repairs) 5881.5556 4841.4554 6921.6557 |
10. | Domestic Repair record 1978(5 repairs) 4204.5 3772.3279 4636.6721 |
+------------------------------------------------------------------------------+
The variable name Price appears unnecessary in the variable out_names. This is remedied by the utility commands -strofnum- and -strtonum-:
strofnum out_names
replace out_names = subinstr(out_names, ", Price", "",.)
strtonum out_names
To create a CI plot
To add estimate and CI as labels a marker label variable is made and some labelling is done:
generate lbl = string(out_mean, "%6.0f") + " (" + string(out_ci95__lb, "%6.0f") + "; " + string( out_ci95__ub, "%6.0f") + ")"
label variable out_eq "Origin"
label variable out_names "Repair record 1978"
Now the graph can be generated:
twoway (scatter out_names out_mean, mlabel(lbl) mlabsize(vsmall) mlabposition(12)) ///
(rcap out_ci95__lb out_ci95__ub out_names, horizontal) ///
, yscale(range(.5 5.5)) ///
ylabel(1(1)5, angle(zero) valuelabel) ///
by(out_eq, legend(off) cols(1) note("")) ytitle(Mean price and 95% CI) name(g1, replace)
To create a slightly different CI plot
An alternative using the option ziprows is:
sysuse auto, clear
label define rep78 1 "1 repair" 2 "2 repairs" 3 "3 repairs" 4 "4 repairs" 5 "5 repairs"
label values rep78 rep78
sumat price if foreign == "Foreign":origin, statistics(mean ci) rowby(rep78) roweq(Foreign) full
matrix out = r(sumat)
sumat price if foreign == "Domestic":origin, statistics(mean ci) rowby(rep78) roweq(Domestic) full
matrix out = out \ r(sumat)
matrix2stata out, clear ziprows
Now the dataset looks like:
list
+------------------------------------------------+
| out_rowe~s out_mean out_ci~lb out_ci~ub |
|------------------------------------------------|
1. | Foreign . . . |
2. | Repair rec . . . |
3. | Repair rec . . . |
4. | Repair rec 4828.6667 3373.8855 6283.4478 |
5. | Repair rec 6261.4444 5022.6874 7500.2015 |
|------------------------------------------------|
6. | Repair rec 6292.6667 4485.8221 8099.5113 |
7. | 7 . . . |
8. | Domestic . . . |
9. | Repair rec 4564.5 3840.2933 5288.7067 |
10. | Repair rec 5967.625 3487.3029 8447.9471 |
|------------------------------------------------|
11. | Repair rec 6607.0741 5226.0617 7988.0865 |
12. | Repair rec 5881.5556 4841.4554 6921.6557 |
13. | Repair rec 4204.5 3772.3279 4636.6721 |
14. | 0 . . . |
+------------------------------------------------+
And the graph is generated by:
generate lbl = string(out_mean, "%6.0f") + " (" + string(out_ci95__lb, "%6.0f") + "; " + string( out_ci95__ub, "%6.0f") + ")"
twoway (scatter out_roweqnames out_mean, mlabel(lbl) mlabsize(vsmall) mlabposition(12)) ///
(rcap out_ci95__lb out_ci95__ub out_roweqnames, horizontal) ///
, ylabel(1(1)6 8(1)13, angle(zero) valuelabel) ///
legend(off) ytitle(Mean price and 95% CI) name(g2, replace)
Last update: 2022-04-19, Stata version 17