In grounded theory, interpretation (in the form of analysis) occurs throughout the data collection process. Initially "open coding" keeps the analysis at a more descriptive level, but as patterns are repeated in further data collection, then you begin to create theoretical categories. From there, it is important to keep comparing your categories to each other as you collect new data. Once the theoretical categories stabilize and are not affected by further data collection, then you have reached saturation.

Of course, you could claim to reach saturation too early (under- interpretation), or you could keep collecting data when there were only minor differences (over-interpretation). But in general, theoretical saturation is the main influence on reaching a reasonable level of interpretation.

In terms of approaches to analysis other than grounded theory, there is a similar "lumper versus splitter" distinction. In other words, some analysts prefer to have fewer and broader categories for their conclusions, even if that ignores some small inconsistencies (these are lumpers). In contrast, other analysts prefer to create more and narrower categories for their results, even if some of those differences are small (these are splitters). Whether this individual tendency amounts to either under- or over-interpretation of the data will depend on personal preferences.

David Morgan provides a useful answer, but I will add to it through simplicity. First, inferences are not fact, and unless observable (even better if said, seen, and artifactual, think Spradley), they must be treated with care. The more abstract and distant, the more interpretive. Secondly, grounded theory relies on "grounded." Grounded means rooted in the data, and a key process is theoretical sampling, which includes testing and checking one's theories (and by extension, inferences).

See the following:

Becker, H., & Geer, B. (1957). Participant observation and interviewing: A comparison. Human Organization, 16(3), 28-32.

Conlon C, Timonen V, Elliott-O’Dare C, O’Keeffe S, Foley G. Confused about theoretical sampling? Engaging theoretical sampling in diverse grounded theory studies. Qualitative Health Research. 2020;30(6):947-959. doi:10.1177/1049732319899139

Thank you for asking this question. I think this is why research design is so important to build in extra triangulation to remedy our biases. Also, getting feedback from other researchers and getting member check would be helpful. And then searching for research related to our findings and interpretations is a must. This is a very relevant and significant question for all of us who use qualitative inquiry.

Essentially (in GT), as for instance, David puts it in his own way, above, constant comparison of incoming and already collected data should prevent overinterpretation; while avoiding a premature 'data saturation' call should prevent under-interpretation.