Fallacies of reification—the placebo effect

TL;DR: I al­ign with the minor­ity po­si­tion that “there is a lot less to the so-called placebo effect than peo­ple tend to think there is (and the name is hor­ribly mis­lead­ing)”, a strong opinion weakly held.

The fol­low­ing post is an off-the cuff re­ply to a G+ post of gw­ern’s, but I’ve been think­ing about this on and off for quite a while. Were I to ex­pand this for post­ing to Main, I would: a) go into more de­tail about the pub­lished re­search, b) in­tro­duce a sec­ond fal­lacy of reifi­ca­tion for com­par­i­son, the so-called “10X var­i­ance in pro­gram­mer pro­duc­tivity”.

My agenda is to have this join my short se­ries of ar­ti­cles on “soft­ware en­g­ineer­ing as a dis­eased dis­ci­pline”, which I view as my mod­est at­tempt at “us­ing Less Wrong ideas in your se­cret iden­tity” and is cov­ered at greater length in my book-in-progress.

I would there­fore ap­pre­ci­ate your feed­back and prob­ing at weak points.

Most of the time, talk of placebo effects (or worse of “the” placebo effect) falls vic­tim to the reifi­ca­tion fal­lacy.

My po­si­tion is roughly “there is a lot less to the so-called placebo effect than peo­ple think there is (and the name is hor­ribly mis­lead­ing)”.

More pre­cisely: the term “placebo” in the con­text of “placebo con­trol­led trial” has some use­ful­ness, when used to mean a par­tic­u­lar way of dis­t­in­guish­ing be­tween the null and test hy­pothe­ses in a trial: namely, that the test and con­trol group re­ceive ex­actly the same treat­ment, ex­cept that you sub­sti­tute, in the con­trol group, an in­ert sub­stance (or in­op­er­a­tive pro­ce­dure) for the pu­ta­tively ac­tive sub­stance be­ing tested.

What­ever out­come mea­sures are used, they will gen­er­ally im­prove some­what even in the con­trol group: this can be due to many things, in­clud­ing re­gres­sion to the mean, the dis­ease run­ning its course, in­creased com­pli­ance with med­i­cal in­struc­tions due to be­ing in a study, ex­pec­tancy effects lead­ing to bi­ased ver­bal self-re­ports.

None of these is prop­erly speak­ing an “effect” causally linked to the in­ert sub­stance (the “placebo pill”). The reifi­ca­tion fal­lacy con­sists of think­ing that be­cause we give some­thing a name (“the placebo effect”) then there must be a cor­re­spond­ing re­al­ity. The false in­fer­ence is “the peo­ple who im­proved in the con­trol group were healed by the power of the placebo effect”.

The fur­ther false in­fer­ence is “there are ail­ments of which I could be cured by in­gest­ing small sugar pills ap­pro­pri­ately la­beled”. Some of my friends ac­tu­ally lev­er­age this into jus­tifi­ca­tion for buy­ing sugar in phar­ma­cies at a ridicu­lous markup. I con­fess to be­ing aghast when­ever this hap­pens in my pres­ence.

A bet­ter name has been sug­gested: the “con­trol re­sponse”. This is ex­per­i­ment-spe­cific, and en­com­passes all of the var­i­ous mechanisms which make it look like “the con­trol group im­proves when given a sugar pill /​ sal­ine solu­tion /​ sham treat­ment”. More­over it avoids hint­ing at mys­te­ri­ous heal­ing pow­ers of the mind.

Meta-analy­ses of those few stud­ies that were de­signed to find an ac­tual “placebo effect” (i.e. stud­ies with a non-treat­ment arm, or stud­ies com­par­ing ob­jec­tive out­come mea­sures for differ­ent place­bos) have not con­firmed it, the few in­di­vi­d­ual stud­ies that find a pos­i­tive effect are in­con­clu­sive for a va­ri­ety of rea­sons.

Doubt­ing the ex­is­tence of the placebo effect will ex­pose you to im­me­di­ate con­tra­dic­tion from your ed­u­cated peers. One ex­pla­na­tion seems to be that the “placebo effect” is a nec­es­sary ar­gu­men­ta­tive prop in the ar­se­nal of two op­posed “camps”. On the one hand pro­po­nents of CAM (Com­ple­men­tary and Alter­na­tive Medicine) will ar­gue that “even if a herbal rem­edy is a placebo, who cares as long as it ac­tu­ally works” and must there­fore as­sume that the placebo effect is real. On the other hand op­po­nents of CAM will say “home­opa­thy or herbal reme­dies only seem to work be­cause of the placebo effect, we can there­fore dis­miss all pos­i­tive re­ports from peo­ple treat­ing them­selves with such”.

I don’t have a proper list of refer­ences yet, but see the fol­low­ing: