Human direct brain recordings have implicated 4-8 Hz theta-band activity in a wide range of cognitive processes including spatial navigation, working memory, and associative recall (Kahana et al. (2001); Nyhus & Curran (2010)). Yet, many highly-powered memory studies indicate that good memory appears during periods of reduced theta and alpha (8-12 Hz) activity and enhanced high-frequency (>40 Hz activity), a pattern known as the spectral tilt (Burkeet al., 2015). In a meta-analysis of prior work, Herweg et al. (2020) suggest several possible resolutions to these discrepant results: They hypothesize that theta increases appear under specific conditions: (1) measures that isolate narrow-band oscillatory effects by filtering out broad-band power fluctuations like spectral tilt, (2) contrasts that isolate retrieval processes, and specifically associative effects during retrieval, (3) measures that aggregate neural activity over larger brain areas, thus picking up synchronous oscillations. We evaluate each of these three hypotheses in a large dataset of human hippocampal depth recordings (674 electrode pairs across 144 patients). Using the IRASA method to separate broad and narrow-band components of the field potential we show that (1) isolating the narrow-band components of the signal does recover an increase in theta power at both encoding and retrieval. (2) theta does not reliably increase as a function of associative strength during retrieval, and (3) employing an average reference scheme - which aggregates neural activity over the entire brain - rather than the more localized bipolar reference scheme does not produce significantly different results.