A quick search on Google Scholar found me the following paper by Amy J. Henley & Florence D. DiGennaro Reed: Should You Order the Feedback Sandwich? Efficacy
of Feedback Sequence and Timing
I am no psychiologist, and can't really judge the research in-depth, but it looks like they put some effort into it, so here goes.
Participants and Setting
Participants were eight undergraduate students (7 females, 1 male) enrolled
in an introductory behavioral science course at a midwestern university who
received extra credit for participation. The experimenter was a graduate
teaching assistant for six of the eight participants. Participants’ ranged in age
from 18 to 43 (M = 23). Experimental sessions took place in a research room
(2.21 × 2.03 × 2.44 m) containing a table, a chair, experimental materials,
and one bin located on the center-right of the table for completed products.
A one-way mirror separated the research room from an observation room of
the same dimensions.
Participants completed four simulated office tasks: folding brochures, stuffing
envelopes, collating packets, and filing timesheets. For the folding task, the
experimenter instructed participants to fold brochures in half and place each
one in the completion bin located on the table. The materials for stuffing
envelopes included two flyers announcing a community event and a box of
500 envelopes. The experimenter instructed participants to place one of each
flyer in an envelope and place the unsealed envelope in the completion bin.
Materials for the third simulated office task, collating packets, included seven
pages of a training manual. The experimenter positioned stacks of each page
in two horizontal rows centered in front of the participant (four stacks on the
top row, three on the bottom) and a stapler in the bottom-right open space.
The experimenter instructed participants to gather one page from each stack,
staple the packet in the corner, and place the packet in the completion
bin. For the remaining task, we created 120 timesheets (four timesheets for
30 employees). The timesheets were pseudo-randomized and placed on the
table aside a mobile bin containing 30 hanging files, one for each employee.
We grouped files alphabetically by first name. The experimenter instructed
participants to identify the name on the timesheet and file the timesheet in
the corresponding folder.
The three sequences of feedback included (a) the feedback sandwich, or
the delivery of a positive statement followed by a corrective statement
and another positive statement (PCP); (b) a positive–positive–corrective
(PPC) sequence; and (c) a corrective–positive–positive (CPP) sequence.
We selected these sequences to hold the ratio of positive to corrective statements constant and only vary the delivery sequence.1 We also evaluated
the effects of no feedback as a control condition. Four participants received
feedback about their prior performance immediately before completing the
next session of the same task (i.e., presession feedback). The remaining
four participants received feedback immediately after the completion of each
session (i.e., postsession feedback).
On arriving for the first session, participants provided informed consent and
demographic information. We asked them to refrain from using their mobile
devices during sessions. The experimenter presented the materials on the
tabletop, provided instructions about how to perform each task, and asked
participants to complete the task. During each session, the experimenter
observed the participant through a one-way mirror. After 5 min, the experimenter
knocked on the window to prompt the participant to stop performing
the task, entered the research room, gathered the session materials, and
began the next session. Participants did not receive feedback for any of the
tasks. Baseline continued until the participant completed each of the four
tasks a minimum of three times and the rate of performance was stable.
Their conclusions were as follows:
The purpose of this study was to evaluate the efficacy of feedback
sequence—in particular the feedback sandwich method—and the influence
of the timing of feedback delivery. Overall, the findings suggest that the
sequence of feedback statements and their timing influence performance,
but the effects may be idiosyncratic across participants at the individual
level. However, interesting findings emerge when we compare aggregate
performance. For participants who experienced presession feedback, the no
feedback condition was the most efficacious and the PPC sequence was
the least efficacious. For participants who received postsession feedback,
the CPP sequence was the most efficacious and the no feedback condition
was the least efficacious. Although the most and least efficacious feedback
sequences differed for the pre- and postsession feedback conditions when
we considered all conditions, there were no statistically significant differences
in performance based on feedback timing within a particular feedback
sequence, except for the no feedback condition.
More details and more elaborate discussion can be found in the link above.
So to answer your question, there is at least some research done to test the validity of the feedback sandwich, but it doesn't seem all too convincing there is a real benefit.