apop_OLS.c File Reference

Functions

void * apop_ls_settings_copy (apop_ls_settings *in)
void apop_ls_settings_free (apop_ls_settings *in)
apop_ls_settings * apop_ls_settings_init (apop_ls_settings in)

Variables

Detailed Description

OLS models. Much of the real work is done in apop_regression.c.

Function Documentation

apop_ls_settings* apop_ls_settings_init ( apop_ls_settings in )

Initialize the settings for a least-squares--type model. For use with Apop_model_add_group. See apop_ls_settings for the possible elements to set.

Variable Documentation

apop_model apop_iv

Instrumental variable regression

Operates much like the apop_ols model, but the input parameters also need to have a table of substitutions. The vector element of the table lists the column numbers to be substituted (the dependent var is zero; first independent col is one), and then one column for each item to substitute.

If the vector of your apop_data set is NULL, then I will use the row names to find the columns to substitute. This is generally more robust and/or convenient.

If the instruments data set is somehow NULL or empty, I'll just run OLS.

apop_data *submatrix =apop_data_alloc(0, data->matrix->size1, 2);
APOP_COL(submatrix, 0, firstcol);
gsl_vector_memcpy(firstcol, your_data_vector);
APOP_COL(submatrix, 1, secondcol);
gsl_vector_memcpy(firstcol, your_other_data_vector);
apop_name_add(submatrix->names, "subme_1", 'r');
apop_name_add(submatrix->names, "subme_2", 'r');

Apop_model_add_group(&apop_iv, apop_ls, .instruments = submatrix);
apop_model *est = apop_estimate(data, apop_iv);
apop_model_show(est);

Todo:: This function does some serious internal data copying. It would be only slightly more human- and labor-intensive to do the linear algebra without producing the Z matrix explicitly.

apop_model apop_ols

Ordinary least squares.

Parameters:

	inset	The first column is the dependent variable, and the remaining columns the independent. Is destroyed in the process, so make a copy beforehand if you need.
	epin	An apop_model object. It may have a apop_ls_settings group attached. I'll look at the `destroy_data` element; if this is NULL or `destroy_data==0`, then the entire data set is copied off, and then mangled. If `destroy_data==1`, then this doesn't copy off the data set, but destroys it in place. I also look at `want_cov` and `want_expected_value`, though I'll not produce the covariance matrix only if both are `'n'`.

Returns:: Will return an apop_model *. The_result->parameters will hold the coefficients; the first coefficient will be the coefficient on the constant term, and the remaining will correspond to the independent variables. It will therefore be of size (data->size2). Do not pre-allocate.

If you asked for them, the covariance matrix, confidence levels, and residuals will also be returned. The confidence intervals give the level of certainty with which we can reject the hypothesis that the given coefficient is zero.

See also the page on Data prep rules.

sample

First, you will need a file named data in comma-separated form. The first column is the dependent variable; the remaining columns are the independent. For example:

Y, X_1, X_2, X_3
2,3,4,5
1,2,9,3
4,7,9,0
2,4,8,16
1,4,2,9
9,8,7,6

The program:

#include <apop.h>

int main(void){
apop_data       *data;
apop_model   *est;
    apop_text_to_db("data","d");
    data = apop_query_to_data("select * from d");
    est  = apop_estimate(data, apop_ols);
    apop_model_show(est);
    return 0;
}

If you saved this code to sample.c, then you can compile it with

gcc sample.c -lapophenia -lgsl -lgslcblas -lsqlite3 -o run_me

and then run it with ./run_me. Alternatively, you may prefer to compile the program using a Makefile .

Feeling lazy? The program above was good form and demonstrated useful features, but the code below will do the same thing in two lines:

#include <apop.h>
int main(){ apop_model_show(apop_estimate(apop_text_to_data("data"), apop_ols)); }

apop_model apop_wls

The WLS model