The omission test is a variant of the substitution test. An example:

In coding a situation where an animate being is affected on one of its body parts, German sometimes has an alternative of two case patterns:

.a.Linda trat mir auf den Fuß.
b.Linda trat mich auf den Fuß.
GermanLinda trod on my foot.

The person affected is in the dative in #a, but in the accusative in #b. Since the two versions are essentially synonyomous, this appears superficially to be an inconsequential free variation in the case of some verbal dependent. The omission test, however, reveals that two different constructions are involved:

.a.*Linda trat mir.
b.Linda trat mich.
GermanLinda kicked me.

This suggests that the accusative constituent in the #b version may be a direct object, while the dative constituent in the #a version probably is not an indirect object, because then it would be inexplicable that the prepositional phrase is inomissible.

Applying the omission test, instead, to the two variant expressions, we get .

.Linda trat auf den Fuß.

This is semantically odd, but grammatical. The semantic oddity disappears if Fuß is replaced by Teppich ‘carpet’. This shows that .b illustrates one valency frame of treten ‘trod, kick’, and exemplifies another one.

If this is so, then .a must be an expanded variant of , featuring a dativus incommodi which triggers the inference of a possessive relation: mir in .a is a so-called external possessor.

Result of the omission test: Far from featuring case variation in a given verb dependent, represents two entirely different constructions:

  1. .a exemplifies a valency frame of treten which includes a directional complement. If this is a body part term, its possessor is commonly coded as a dativus incommodi.
  2. .b exemplifies a valency fram of treten which includes a direct object. This is compatible with an optional directional adjunct.