JSON This

This package provides wrappers around Stata commands to save their results in JSON files. A generic wrapper, json_this, works with any r-class or e-class command. In addition to this, the package also contains three specialized wrappers around:

• reg
• reghdfe
• ttest

The resulting JSON files can be used, for example, to generate arbitrarily formatted regression tables using the coeftable Python package.

Installation

capture ado uninstall json_this
net install json_this, from("https://raw.githubusercontent.com/gn0/json-this/master/ado/")

Usage

json_reg and json_reghdfe

The usage of json_reg is nearly identical to that of json_reghdfe. This subsection illustrates the latter.

Save the results of reghdfe y x, absorb(z) in a file called foo.json:

json_reghdfe y x using foo.json, absorb(z)

foo.json looks something like this (broken up here into multiple lines for legibility):

{"outcome": "y",
 "N": 300,
 "N_clust": null,
 "clustvar": "",
 "absvars": "a",
 "df_m": 1,
 "df_r": 249,
 "F": 40.68007172613321,
 "F_p": 8.66178084844e-10,
 "r2": 0.2923736048196345,
 "r2_within": 0.1404310330487373,
 "r2_a": 0.1502799511689586,
 "r2_a_within": 0.1369789488441137,
 "mean_outcome": 0.2714693066834782,
 "coef": {"x": {
            "est": 0.4606558396069469,
            "se": 0.0722246965709874,
            "t": 6.378093110494174,
            "p": 8.66178084844e-10,
            "stars": "***",
            "ci_l": 0.3184066369562497,
            "ci_u": 0.6029050422576441}}}

The replace option allows Stata to overwrite foo.json if it already exists:

json_reghdfe y x using foo.json, absorb(z) replace

if conditions, weights, and other options can also be passed on to reghdfe:

json_reghdfe y x using foo.json if a == 1, absorb(z) replace
json_reghdfe y x using foo.json [aweight=w], absorb(z) replace
json_reghdfe y x using foo.json, absorb(z) vce(robust) replace

json_ttest

Save the results of ttest x == z in a file called foo.json:

json_ttest x == z using foo.json

foo.json looks similar to the following (again, broken up here into multiple lines for legibility):

{"spec": "x == z",
 "stars": "",
 "stars_l": "",
 "stars_u": "",
 "N_1": 300,
 "N_2": 300,
 "p_l": 0.77231319,
 "p_u": 0.22768681,
 "p": 0.45537361,
 "se": 0.08032392,
 "t": 0.74745958,
 "sd_1": 1.0311098,
 "sd_2": 1.012639,
 "sd": null,
 "mu_1": 0.01498158,
 "mu_2": -0.0450573,
 "df_t": 299,
 "if_cond": "",
 "by_var": "",
 "diff": 0.06003889}

The replace option can be used again to allow Stata to overwrite foo.json:

json_ttest x == z using foo.json, replace

if conditions and other options can also be passed on to ttest:

json_ttest x == z if a == 1 using foo.json, replace
json_ttest x using foo.json, by(a) replace
json_ttest x == z using foo.json, unpaired replace

json_this

json_reg, json_reghdfe, and json_ttest are specifically designed to export results from linear regressions and hypothesis tests. Results from other types of commands can be exported with json_this. json_this is a generic wrapper for r-class and e-class commands.

Save the results of summarize x, detail in a file called foo.json:

json_this summarize x using foo.json, detail

foo.json looks similar to the following (broken up into multiple lines for legibility):

{"cmd": "summarize",
 "r_scalars": {
   "N": 15805,
   "sum_w": 15805,
   "mean": 768.3155963302752,
   "Var": 27148076.45847922,
   "sd": 5210.381603921081,
   "skewness": 55.50313309587089,
   "kurtosis": 3639.847818166637,
   "sum": 12143228,
   "min": 7,
   "max": 362289,
   "p1": 27,
   "p5": 52,
   "p10": 77,
   "p25": 137,
   "p50": 270,
   "p75": 551,
   "p90": 1138,
   "p95": 2114,
   "p99": 9671}}

Save the results of lincom, after running reg, in a file called bar.json:

reg y x z
json_this lincom 3 * x - 10 * z using bar.json

json_this does not clear r() and e(). If it did, it would not be able to run lincom. One side effect of this is that bar.json contains results from both lincom and reg:

{"cmd": "lincom",
 "r_scalars": {
   "df": 12775,
   "se": 510.7233517176288,
   "estimate": 306.5145980089461},
 "e_scalars": {
   "rank": 3,
   "ll_0": -125549.2076105619,
   "ll": -107910.0624743788,
   "r2_a": 0.9367528438844114,
   "rss": 16188712291.35621,
   "mss": 239810888673.0277,
   "rmse": 1125.707858761062,
   "r2": 0.9367627440419,
   "F": 94620.99417362853,
   "df_r": 12775,
   "df_m": 2,
   "N": 12778},
 "e_macros": {
   "cmdline": "regress y x z",
   "title": "Linear regression",
   "marginsok": "XB default",
   "vce": "ols",
   "depvar": "y",
   "cmd": "regress",
   "properties": "b V",
   "predict": "regres_p",
   "model": "ols",
   "estat_cmd": "regress_estat"}}

json_this can also be used to export results from user-written programs. For example, the following program called baz saves its results in r() scalars:

program define baz, rclass
    return scalar a = 2
    return scalar b = 3
    return scalar c = 5
end

Save these in a file called baz.json:

json_this baz using baz.json

Then baz.json contains:

{"cmd": "baz",
 "r_scalars": {
   "c": 5,
   "b": 3,
   "a": 2}}

json_this also accepts the replace option to allow Stata to overwrite the output file.

Author

Gabor Nyeki. Contact information is on http://www.gabornyeki.com/.

License

This package is licensed under the Creative Commons Attribution 4.0 International License: http://creativecommons.org/licenses/by/4.0/.