As russhh mentioned, different probes will yield different baseline signals. Some may simply hybridize more easily to their targets or their target sequences may be more abundant because of some bias during the fragmentation of the transcripts etc. That's why you need to look at the relative values for every probe across all samples to get a feeling for whether there might be a systematic reduction associated with individual sample (groups).

You need to remember that, in contrast to RNA-seq, where you're actually sequencing "real" transcripts, microarrays yield a much more indirect measure of transcript levels, which is one of the reasons they've been somewhat overtaken by RNA-seq despite their relative cost efficiency for some applications.

I highly recommend reading up on this if you really want to get to terms with this, e.g. Held et al., Wang et al., or a more recent write-up.