In many cases, even through the early 2000's, publishers produced print journals using older type setting systems and did not produce PDFs of the papers for online publication. Once the papers were type set for printing, the publisher might have discarded all of the files associated with the papers or they might have just kept a file containing a print-ready image of the paper.
Many journals had old material of this sort dating back decades. Since it would take a huge effort to type-set all of those older papers again, journal publishers have settled for simply scanning the earlier papers (or retrieving saved images of the papers) and then publishing the images of the papers online.
Using optical character recognition (OCR) is one way to extract the text from these images. However, OCR isn't 100% accurate, and in any case it can't preserve all of the formatting of the paper. Thus OCR text from a scanned image of a published paper may be somewhat useful for indexing, but it really isn't equivalent to having the original document as a text PDF.
A different issue is that you may have received a copy of a paper from someone other than the publisher. For example, if you made an inter library loan request to get a copy of a paper, it's possible that someone at another university library scanned a printed copy of the paper and sent that scan to you.