To understand the issue it is necessary to understand the process of adding and removing pulldown to 24 frame footage, as well as how 4:2:0 subsampling works. First, a brief refresher on pulldown:

If you expose 24fps footage (e.g. on a film camera) and transfer it to NTSC video, it is necessary to turn 24 images into 30 video frames, each of which consists of 2 fields. This is done by adding what is know as 2:3 pulldown. It turns your first 4 frames of film:

| A | B | C | D |

into 5 frames of video by repeating certain fields like this (each letter here represents one field of video):

| AA | BB | BC | CD | DD |

at which point the sequence repeats itself. In order to get back to 24 frames you need to pull out 2 fields for each original frame. The A, B and D frames are easy (since there are already complete frames for those), but the C frame must be reconstructed from the second field of the BC frame and the first field of the CD frame. This is where the 4:2:0 issue comes in.

Many (some?) people are familiar with 4:2:2 and 4:1:1 subsampling, where the frequency of chroma samples is 50% and 25% respectively of that of luma. All subsampling in those cases is horizontal: every field is subsampled identically. In 4:2:0 subsampling, however, chroma is subsampled by 50% both horizontally _and_ vertically. Vertical subsampling is more complicated that horizontal subsampling. With progressive frames it is fairly straightforward: samples are arranged like this:

|C| |C| |
| | | | |
|C| |C| |
| | | | |

For interlaced footage, however, a different scheme is needed or else one field will have chroma information from the preceding field (and out of sync with the Luma). In this case the field from each field is subsampled separately, like this (black lines are from the odd field, red from the even):

|C| |C| |
|C| |C| |
| | | | |
| | | | |

The problem of properly encoding 2:3 pulldown is now becoming apparent: Ideally, since we wish to end up with progressive frames, we would use the progressive form of vertical subsampling. This would make it impossible (or at least extremely difficult) to remove the pulldown, since there will be cross-frame chroma contamination in the BC and CD frames, making a clean extraction of the C frame difficult. In order to avoid this problem, we need to subsample the video as interlaced video, but this results in significant loss in vertical chroma resolution in the progressive frames, since our chroma samples are now out of order.

Clearly, then, with any format that uses 4:2:0 subsampling ( _cough_ *HDV* _cough_), it is best for the interlaced to stay interlaced and the progressive to stay progressive and to leave it at that.