What is this one-hot-like encoding of ordinal variables called and is it feasible? Common understanding seems to be that when we're doing machine learning (classification or regression) using linear models, there are basically two ways to encode ordinal variables:

*

*Ordinal scale: 1, 2, 3

*One-Hot: [1,0,0], [0,1,0], [0,0,1]

I don't remember where I've learnt this, but this is how someone told me to do it:
|           | v_1 | v_2 | v_3 | v_4 |
-------------------------------------
| very bad  |  0  |  0  |  0  |  0  |
| bad       |  1  |  0  |  0  |  0  |
| neutral   |  1  |  1  |  0  |  0  |
| good      |  1  |  1  |  1  |  0  |
| very good |  1  |  1  |  1  |  1  |

This way we preserve the ordering of the values from "very bad" to "very good" (as long the weights of these dummy variables are non-negative), but we don't imply that the difference between "neutral" and "good" is the same as between "good" and "very good".
I've found only one source in German for this. They are slides from a lecture that call it "Ordinale Entflechtung" (I have no idea on how to translate this, maybe "ordinal disentanglement"?), but there is not a single other source using this term.
Is this a known concept? If yes, what is it called? And is it feasible or something one should avoid?
 A: I know this paper promotes this method and does not seem to credit another source for it, nor name it, though they refer to coding "between-strata differences" in the ordinal variable:
Walter, S. D., Feinstein, A. R., & Wells, C. K. (1987). Coding ordinal independent variables in multiple regression analyses. American Journal of Epidemiology, 125(2), 319-323.
From that paper:

It's certainly possible there is an older reference that the authors were unaware of at the time.
As the authors point out, this coding scheme is algebraically equivalent to other coding schemes (dummy coding, reference coding), but it can be a convenient way to express the outcome:

It should be pointed out that if one has
(k + 1) strata and one parameterizes a full
set of k independent variables, it is always
possible to "convert" from any one coding
scheme to any other, obtaining point estimates and variances; this will typically involve some algebraic manipulation of the
codings, and of the corresponding variancecovariance matrices for the parameters.
However, the overall significance test of
association on k degrees of freedom, comparing differences between all the strata
simultaneously, is the same for any valid
coding scheme.

