help tostregress ------------------------------------------------------------------------------------------------------------------------- Title tostregress -- Linear regression tests for equivalence Syntax tostregress depvar [indepvars] [if] [in] [weight] [, eqvtype(typelist) eqvlevel(numlist) uppereqvlevel(numlist ) relevance level(#) regress_options] options Description ------------------------------------------------------------------------------------------------------------------- Options eqvtype(typelist) specify equivalence threshold with Delta or epsilon for each indepvar eqvlevel(numlist) the level of tolerance defining the equivalence interval for each indepvar uppereqvlevel(numlist) the upper value of an asymmetric equivalence interval for each indepvar level(#) set nominal type I level; default is level(95) coressponding to alpha=0.05 relevance perform & report combined tests for difference and equivalence regress options all the options allowed with regress svy(svy options) all the options allowed with the svy prefix, including vcetype ------------------------------------------------------------------------------------------------------------------- indepvars may contain factor variables; see fvvarlist. depvar and indepvars may contain time-series operators; see tsvarlist. time-series operators may not be combined with factor variable operators. by, and statsby are allowed; see prefix. svy is allowed via the svy() option, but not as a prefix command. bootstrap and jackknife are allowed using vce(), or svy(), but are not allowed as prefix commands. See Remarks. Weights are allowed with neither vce(bootstrap), nor svy(bootstrap). aweights are allowed with neither vce(jackknife), nor svy(jackknife). hascons, tsscons, vce(), beta, noheader, notable, plus, depname(), mse1, and weights are not allowed with svy(). aweights, fweights, iweights, and pweights are allowed; see weight. See [R] regress postestimation for features available after estimation. Description tostregress tests for the equivalence of each regression coefficient and zero within separate symmetric equivalence intervals defined by eqvtype and eqvlevel for using a two one-sided t tests approach (Schuirmann, 1987). Typically ('positivist') null hypotheses are framed from an assumption of a lack of difference between two quantities, and reject this assumption only with sufficient evidence. When performing tests for equivalence, one frames a ('negativist') null hypothesis with the assumption that two quantities are different by at least as much as an equivalence interval defined by some chosen level of tolerance. An equivalence null hypothesis takes one of the following two forms depending on whether equivalence is defined in terms of Delta (equivalence expressed in the same units as the parameter for an independent variable), or in terms of epsilon (equivalence expressed in the units of the T disribution with the given degrees of freedom): Ho: |_b_x| >= Delta, where the equivalence interval ranges from _b_x - Delta to _b_x + Delta, and where _b_x is the parameter being estimated for x. This translates directly into two one-sided null hypotheses: Ho1: Delta - _b_x <= 0; and Ho2: _b_x + Delta <= 0 -OR- Ho: |T| >= epsilon, where the equivalence interval ranges from -epsilon to epsilon. This also translates directly into two one-sided null hypotheses: Ho1: epsilon - T <= 0; and Ho2: T + epsilon <= 0 When an asymmetric equivalence interval is defined using the uppereqvlevel option the general negativist null hypothesis becomes: Ho: _b_x <= Delta_lower, or _b_x >= Delta_upper, where the equivalence interval ranges from _b_x + Delta_lower to _b_x + Delta_upper. This also translates directly into two one-sided null hypotheses: Ho1: Delta_upper - _b_x <= 0; and Ho2: _b_x - Delta_lower <= 0 -OR- Ho: T <= epsilon_lower, or T >= epsilon_upper, Ho1: epsilon_upper - T <= 0; and Ho2: T - epsilon_lower <= 0 The two one-sided test statistics corresponding to Ho1 and Ho2, t1 and t2, are both constructed so that their p-values are upper (right) tail probabilities: p1 = P(T>=t1) p2 = P(T>=t2) NOTE: the appropriate level of alpha implied by level is precisely the same as in the corresponding two-sided test for mean difference, so that, for example, if one wishes to make a type I error %5 of the time, one simply conducts both of the one-sided tests of Ho1 and Ho2 by comparing the resulting p-value to 0.05 (Tryon and Lewis, 2008). Options +------+ ----+ Main +------------------------------------------------------------------------------------------------------- eqvtype(typelist) defines whether the equivalence interval will be defined in terms of Delta or epsilon (delta, or epsilon). These options change the way that evqlevel is interpreted: when delta is specified, the evqlevel is measured in the units of the variable being tested, and when epsilon is specified, the evqlevel is measured in units of the T distribution; put another way epsilon = Delta/standard error. The default is delta. eqvtype must be specified with no entries, one entry, or the same number of entries as parameters (coefficients) being estimated, including the intercept term, if any. The below examples are for a model with three parameter estimates -- one for weight, one for foreign, and one for _cons (the intercept term): The first example for eqvtype gives the default equivalence type (delta) either omitting the option (as shown here) or by leaving it empty): . tostregress mpg weight foreign, eqvlevel(5) This example for eqvtype gives a single equivalence type to be used for all three parameters: . tostregress mpg weight foreign, eqvtype(epsilon) eqvlevel(2.145) The last example for eqvtype gives a separate equivalence type for each parameter: . tostregress mpg weight foreign, eqvtype(delta delta epsilon) eqvlevel(5 5 2.145) Defining tolerance in terms of epsilon means that it is not possible to reject any test for mean equivalence Ho if epsilon <= the critical value of t for a given level and degrees of freedom. Because epsilon = Delta/standard error, we can see that for the same reason it is not possible to reject any Ho if Delta <= the product of the standard error and critical value of t for a given level and degrees of freedom. tostregress reports when either of these conditions obtain. eqvlevel(numlist) defines the equivalence threshold for the tests depending on whether eqvtype is delta or epsilon (see above). Researchers are responsible for choosing meaningful values of Delta or epsilon. The default value for each coefficient is 1 when delta is the eqvtype for that coefficient, and 2 when epsilon is the eqvtype for that coefficient. eqvlevel must be specified with no entries, one entry, or the same number of entries as parameters (coefficients) being estimated, including the intercept term, if any. uppereqvlevel(#) defines the upper equivalence threshold for the test, and transforms the meaning of eqvlevel to mean the lower equivalence threshold for the test. Also, eqvlevel is assumed to be a negative value. Taken together, these correspond to Schuirmann's (1987) asymmetric equivalence intervals. If uppereqvlevel==|eqvlevel|, then uppereqvlevel will be ignored. eqvtlevel must be specified with no entries, one entry, or the same number of entries as parameters (coefficients) being estimated, including the intercept term, if any. The below examples are for a model with three parameter estimates -- one for weight, one for foreign, and one for _cons (the intercept term): level(#) specifies the nominal type I error rate. The default is level(95), which corresponds to alpha = 0.05. relevance reports results and inference for combined tests for difference and equivalence for specific level, eqvtype, and eqvlevel choices. See the Remarks section more details on inference from combined tests. svy(svy options) estimates the model as if with the svy prefix, and any svy options you include. You may also specify vcetype options for svy here. See Remarks about jackknife and bootstrap options for svy. Remarks As described by Tryon and Lewis (2008), when both tests for difference and equivalence are taken together, there are four possible interpretations: 1. One may reject the null hypothesis of no difference, but fail to reject the null hypothesis of difference, and conclude that there is a relevant difference between _b_x and zero as large as Delta or epsilon. 2. One may fail to reject the null hypothesis of no difference, but reject the null hypothesis of difference, and conclude that _b_x is equivalent to zero within the equivalence range (i.e. defined by Delta or epsilon). 3. One may reject both the null hypothesis of no difference and the null hypothesis of difference, and conclude that _b_x is trivially different, from zero within the equivalence range (i.e. defined by Delta or epsilon). 4. One may fail to reject both the null hypothesis of no difference and the null hypothesis of difference, and draw an indeterminate conclusion, because the data are underpowered to detect difference or equivalence for _b_x and zero. Caveat Emptor: jackknife and bootstrap options for svy estimation have been implemented only at a basic level. If you run into problems with these options, especially if using suboptions for either estimator, please share your syntax and data with me so that I may improve tostregress. Examples These examples correspond to those written in the help file for regress: Setup . sysuse auto Report equivalence tests for a linear regression . tostregress mpg weight foreign, eqvtype(epsilon) eqvlevel(2.145) Report equivalence tests for a linear regression, but add a relevance test report . tostregress mpg weight foreign, eqvtype(epsilon) eqvlevel(2.145) relevance Fit a better linear regression, from a physics standpoint, but add asymmetric intervals, and report equivalence and relevance tests . gen gp100m = 100/mpg . tostregress gp100m weight foreign, eqvtype(epsilon) eqvlevel(2.145) upper(1.895) rel Obtain beta coefficients without refitting model . tostregress, beta eqvtype(epsilon) eqvlevel(2.145) upper(1.895) rel Report equivalence tests when suppressing the intercept term . tostregress weight length, noconstant eqvtype(delta) eqvlevel(5) Report equivalence tests when the model already has constant . tostregress weight length bn.foreign, hascons eqvtype(delta epsilon epsilon) eqvlevel(5 2.145 2.145) Examples: equivalence and relevance tests for regression with robust standard errors ----------------------------------------------------------------------------------------------------------------- . sysuse auto, clear . generate gpmw = ((1/mpg)/weight)*100*1000 . tostregress gpmw foreign, eqvtype(epsilon) eqvlevel(2.145) rel . tostregress gpmw foreign, vce(robust) eqvtype(epsilon) eqvlevel(2.145) rel . tostregress gpmw foreign, vce(hc2) eqvtype(epsilon) eqvlevel(2.145) rel . tostregress gpmw foreign, vce(hc3) eqvtype(epsilon) eqvlevel(2.145) rel ----------------------------------------------------------------------------------------------------------------- . webuse regsmpl, clear . tostregress ln_wage age c.age#c.age tenure, vce(cluster id) eqvt(epsilon) eqvl(2.145) ----------------------------------------------------------------------------------------------------------------- Example: equivalence tests for weighted regression . sysuse census . tostregress death medage i.region [aw=pop], eqvtype(epsilon) eqvlevel(1.782) level(90) Examples: equivalence tests for linear regression with survey data Setup . webuse highschool Perform linear regression using survey data . svy: regress weight height Perform corresponding linear regression tests for equivalence using survey data . tostregress weight height, eqvt(epsilon) eqvl(2.145) rel svy() Setup . generate male = sex == 1 if !missing(sex) Perform linear regression using survey data for a subpopulation . svy, subpop(male): regress weight height Perform corresponding linear regression tests for equivalence using survey data for a subpopulation . tostregress weight height, eqvt(epsilon) eqvl(2.145) rel svy(subpop(male)) Perform the above survey data estimate using the jackknife variance estimator . tostregress weight height, eqvt(epsilon) eqvl(2.145) rel svy(jackknife subpop(male)) Saved results In addition to the information saved by regress, tostregress saves the following in e(): Scalars e(alpha) The nominal Type I error rate based on the level option (multiple comparisons issues will arise in a multiple regression context). Macros e(cmd) tostregress e(cmdline) canonical form of command as typed e(title) title of estimation output reflecting the svy option e(eqvtype) the eqvtype for each coefficient/test e(lowereqvlevel) the lower eqvlevel for each coefficient/test e(uppereqvlevel) the uppereqvlevel for each coefficient/test e(rel_conclusions) relevance test conclusions if the relevance option is used Matrices e(T1) t1 test statistics e(T2) t2 test statistics e(P1) p-values corresponding to the test statistics in T1 e(P2) p-values corresponding to the test statistics in T2 Author Alexis Dinno Portland State University alexis.dinno@pdx.edu Development of tost is ongoing, please contact me with any questions, bug reports or suggestions for improvement. Fixing bugs will be facilitated by sending along: (1) a copy of the data (de-labeled or anonymized is fine), (2) a copy of the command used, and (3) a copy of the exact output of the command. Suggested citation Dinno A. 2017. tostregress: Linear regression tests for equivalence. Stata software package. URL: https://www.alexisdinno.com/stata/tost.html References Schuirmann, D. A. 1987. A comparison of the two one-sided tests procedure and the power approach for assessing the equivalence of average bioavailability. Pharmacometrics. 15: 657-680 Tryon, W. W., and C. Lewis. 2008. An inferential confidence interval method of establishing statistical equivalence that corrects Tryon’s (2001) reduction factor. Psychological Methods. 13: 272-277 Also See Help: tost, pkequiv, regress