Discrepancy in SIGMOS results

[mgrzywa-logi]

I wanted to ask about discrepancy in SIGMOS results based on a simple experiment I took recently.

I got a clear speech data from ETSI 103 106 standard (Annex C) - 4 talkers (2 Males/2 Females), 20 sentences, each sample is 4 s duration.
Document: https://www.etsi.org/deliver/etsi_ts/103100_103199/103106/01.06.01_60/ts_103106v010601p.pdf
Audio file:
https://docbox.etsi.org/stq/Open/TS%20103%20106%20Wave%20files/Annex_C_Dynastat%20Speech%20Data/Dyna-Src_P835_16_sentences_4convergence.wav

I cut this audio file for 20 separated, 4 seconds files per every utterance and calculated SIGMOS for every file and for the original 80 seconds file.
What I expected is that the average results of very single utterance will correlate with a result of a concatenated file, but actually they differ as presented below:

<style> </style>

FILENAME 1	COL	DISC	LOUD	NOISE	REVERB	SIG	OVRL	COMMENTS
convergance_utt_1	3,93	4,41	4,36	4,61	4,66	4,22	3,87	male
convergance_utt_2	3,78	4,71	3,58	4,45	4,71	3,94	3,75	female
convergance_utt_3	4,08	4,60	4,27	4,54	4,66	4,31	4,05	male
convergance_utt_4	3,75	4,32	3,85	4,44	4,75	3,91	3,46	female
utt_1	4,04	4,46	4,06	4,61	4,56	3,81	3,55	male
utt_10	3,68	4,87	4,24	4,70	4,93	4,40	3,99	female
utt_11	3,64	4,44	3,59	4,32	4,59	4,04	3,61	male
utt_12	4,30	4,42	3,96	4,53	4,79	4,18	3,70	female
utt_13	3,65	4,31	4,08	4,79	4,63	3,43	3,17	male
utt_14	3,69	4,62	3,68	4,00	4,77	3,83	3,49	female
utt_15	3,65	3,52	3,77	4,37	4,52	3,36	3,05	male
utt_16	3,59	4,33	4,24	4,17	4,55	3,69	3,36	female
utt_2	3,72	4,21	3,59	4,54	4,53	3,49	3,22	female
utt_3	3,91	4,91	4,21	4,61	4,88	4,06	3,78	male
utt_4	3,88	4,46	3,52	4,48	4,82	4,17	3,74	female
utt_5	3,67	4,43	4,33	4,67	4,68	3,77	3,36	male
utt_6	3,10	4,32	4,22	4,46	4,82	3,41	2,99	female
utt_7	4,02	4,71	3,30	4,71	4,59	4,54	3,97	male
utt_8	3,83	4,85	4,11	4,57	4,76	3,71	3,57	female
utt_9	4,03	4,88	4,31	4,70	4,44	4,10	3,90	male
Dyna-Src_P835_16_sentences_4convergence	4,22	4,47	4,50	4,78	4,95	4,25	4,05

<style> </style>

	COL	DISC	LOUD	NOISE	REVERB	SIG	OVRL
avgerage	3,797	4,489	3,963	4,515	4,682	3,918	3,578
original	4,223	4,474	4,498	4,780	4,952	4,248	4,046
diff	0,426	-0,015	0,536	0,265	0,269	0,330	0,468
std	0,249	0,315	0,326	0,191	0,132	0,339	0,315

As you can see, the differences can equals almost 0,5 MOS in some cases.

I wonder what can be the reason of this discrepancy.