Face detection is only so accurate. Facial expressions will slightly change how the face scales.  Mode 1 scales by the face.

If you have feet in the cutouts then modes 2, 3 and 4 will scale by the full body and keep much more accurate scaling than mode 1. So if there are feet in the cutouts then you could use mode 2, 3 or 4.  Just scale and place the feet off canvas below the bottom and save the template that way.  Also, if the ball goes above the head sometimes in the poses then use mode 2, not 3 or 4.  Mode 2 scale from center of face to the bottom of the cutout. Modes 3 and 4 scale from top of the cutout to the bottom of the cutout.

However, if you don’t have feet in the cutouts then modes 2, 3 and 4 can have much more variation because they scale to the bottom of wherever you cropped in camera so if that varies then the variation carries over into SPA.

Mode 1 has the least accurate scaling because the entire body is scaled just by how the face detection see the face.  The variations in you examples are perfectly normal for mode 1 and there is no way to improve on this.  So if you need more accurate than what you are posting then shoot the images with feet and use modes 2, 3 or 4.

See this video for comparison of mode 1 vs mode 2.

Here is a link to tutorials for all of the modes and how they work.

Sports Photo Automation Image Alignment Modes