According to Matthews et al. (Environ. Sci. Technol. 2008, 42, 5839–5842), who use an Input-Output Life Cycle Assessment method:
In a more detailed analysis, the same authors use a similar approach and produce the following table [with the last 8 bars comprising service industries (Environ Sci Technol. 2009 Nov 15;43(22):8509-15.):
A few observations: 1) S1 for power generation is very high, as expected; 2) agriculture, mining, and transportation services also have high S1; 3) S2 appears to be higher on average in services than in manufacturing ; 4) upstream supply chain S3 (green) rivals “other” S3, both of which dominate S3 commute, except in education sector; and 5) note that this is not “all” emissions, just proportion of studied emissions.
The authors conclude that:
We recommend protocol organizations to actively make more speciﬁc Scope 3 guidelines available for their constituents by developing sector-speciﬁc protocols for as many carbon intensive sectors as they feasibly can and create broader industry-speciﬁc protocols for other relatively less carbon intensive sectors.