advice please=#could not calculate numerical derivatives missing values...in polychoric ans factor mat.

[Aline Philibert]

hello to all,

I wanted to do a polychoric tests between some binary variables, 4 ordinal variables, and 3 continuous ones ( ny N is 80-small)

polychoric ( cn_v_sen_p cn_v_sen_t0 cn_v_sen_temp cn_v_mast cn_vii_mot cn_vii_jaw cn_ix_tong cn_ix_st cn_ix_ab_mov_tremor cn_ix_ab_mov_fasciculation cn_ix_gag_2 cn_ix_pal cn_ix_art cn_xi_mot oral_score laryngo_score total_score )

#could not calculate numerical derivatives
missing values encountered
numerical derivatives are approximate
nearby values are missing#

I I do not have missing values
I continued...

display r(sum_w)
global N = r(sum_w)
matrix r = r(R)
factormat r, n($N) factors(3)

error after factormat
factormat r, n($N) factors(3)
matrix r has missing values

imagine I do not want to test polychoric relations but directly factor mat....how can I do please, thank you Aline

[George Ford]

I think that can happen when the data isn't rich enough (you don't much uniqueness across all the variables). That's a big ask for 80 observations.

[Aline Philibert]

Dear George, thank you for answering my questions. You are right these data (answers from a questionnaire) are somehow correlated, but we would like to know how on a map (neurological answers about dysphagia/swalloing) and better understand how they are related to other neurological cranial scores. we are working on a small indigenous community n=72. do you have any idea of how to select some variables among the other ones? thank you Aline

[George Ford]

Start by comparing means, some simple graphs, etc.

What is the X of primary interest (and what type of variable is it) and what is the Y of primary interest (and what type of variable is it)?

Are "other cranial scores" continuous, dichotomous, ordered?

[Mike Lacy]

I'm quite sympathetic to George's suggestions to start with something more simple. Presuming that leads to a desire to do a factor analysis of the correlations among the variables, another suggestion here would be to use something other than polychoric correlations. Estimation of polychoric correlations, in my experience, can be a bit brittle, and it rests on assumptions (e.g., underlying Gaussian distributions) that may be unattractive. One alternative would be to use Kendall's Tau-b, which is defined for binary, ordinal, and continuous variables, as it uses only the order of the data values.

ktau cn_v_sen_p cn_v_sen_t0 cn_v_sen_temp cn_v_mast cn_vii_mot cn_vii_jaw ///
   cn_ix_tong cn_ix_st cn_ix_ab_mov_tremor cn_ix_ab_mov_fasciculation cn_ix_gag_2 ///
   cn_ix_pal cn_ix_art cn_xi_mot oral_score laryngo_score total_score, ///
   stats(taub)
mat R = r(Tau_b)
local N = r(N) // you want the N on which the correlations were based
factormat R, n(`N') factors(3)

Note, by the way, that the useful user-written -polychoric- command can be found with -search polychoric-.