I looked into the references from the other answers (and some more references), here is what I learned.
The TL;DR version:
- Levonorgestrel is not effective at preventing pregnancy in women who have already ovulated.
- The product labeling does not reflect evidence showing that levonorgestrel does prevent implantation by some identifiable mechanism; only uncertainty about whether or not it could.
- The main mechanism by which it was once thought that levonorgestrel might prevent implantation is by affecting the endometrium, but studies later suggest that levonorgestrel has no meaningful effect on endometrial receptivity.
- It is not clear whether levonorgestrel has any effect on pregnancy if administered shortly before ovulation, and if it does, it is not clear by what mechanism.
1. Levonorgestrel is not effective at preventing pregnancy in women who have already ovulated.
The main piece of evidence implying that levonorgestrel does not prevent implantation is that Plan B is not effective at preventing pregnancy in women who have already ovulated. For example,
In "Effectiveness of levonorgestrel emergency contraception given before or after ovulation — a pilot study", 2007 :
Among 17 women who had intercourse in the fertile period of the cycle and took the ECP after ovulation occurred (on Days +1 to +2), we could have expected three or four pregnancies; three were observed. Among 34 women who had intercourse on Days −5 to −2 of the fertile period and took ECP before or on the day of ovulation, four pregnancies could have been expected, but none were observed.
In "Contraceptive efficacy of emergency contraception with levonorgestrel given before or after ovulation", 2010 :
Among the 103 women who took LNG-EC before ovulation (days −5 to −1), 16 pregnancies were expected and no pregnancy occurred (p<.0001). Among the 45 women who took LNG-EC on the day of ovulation (day 0) or thereafter, 8 pregnancies occurred and 8.7 were expected (p=1.00).
2. The product labeling does not reflect evidence showing that levonorgestrel does prevent implantation by some identifiable mechanism; only uncertainty about whether or not it could.
Why did anybody ever think that Plan B inhibits implantation of fertilized eggs?
The labeling of Plan B (which in the US, must be approved by the FDA) says:
Plan B One-Step is believed to act as an emergency contraceptive principally by preventing ovulation or fertilization (by altering tubal transport of sperm and/or ova). In addition, it may inhibit implantation (by altering the endometrium). It is not effective once the process of implantation has begun.
Many point to this as evidence that Plan B inhibits implantation.
However, it's not clear that there was ever any direct evidence that levonorgestrel affects implantation. The New York Times claims that at the time that Plan B was going through FDA approval,
Experts say implantation was likely placed on the label partly because daily birth control pills, some of which contain Plan B’s active ingredient, appear to alter the endometrium, the lining of the uterus into which fertilized eggs implant.
Implantation also likely wound up on the label because of what Dr. Gemzell-Danielsson called wishful thinking by some scientists, who thought that if it could also block implantation, it would be even better at preventing pregnancy.
In 2005, when the FDA reviewed Plan B in order to make it available without prescription in some cases, the manufacturer, Teva Pharmaceuticals, asked for the statement about implantation to be removed from the label. According to the NYT again,
Addressing the issue in a 2005 memorandum, Dr. Steven Galson, director of the F.D.A.’s Center for Drug Evaluation and Research, wrote that studies “conclusively demonstrate” that Plan B’s ability to block ovulation, is “responsible for most, if not all, instances in which emergency contraception prevents pregnancy.” But he also said that studies at that time could not exclude the possibility the pills impeded implantation “in a small percentage of women.” He declined a request to be interviewed.
3. The main mechanism by which it was once thought that levonorgestrel might prevent implantation is by affecting the endometrium, but studies later suggest that levonorgestrel has no meaningful effect on endometrial receptivity.
In both in vitro studies and studies on humans, levonorgestrel has no effect on endometrial receptivity when administered in an emergency contraception dose.
From "Effect of levonorgestrel and mifepristone on endometrial receptivity markers in a three-dimensional human endometrial cell culture model", 2009 :
This in vitro model expresses progesterone-regulated endometrial receptivity factors seen in the physiologic condition. Treatment with levonorgestrel did not affect the expression of these endometrial receptivity markers in contrast to mifepristone.
From "A single midcycle dose of levonorgestrel similar to emergency contraceptive does not alter the expression of the L-selectin ligand or molecular markers of endometrial receptivity", 2010 :
To examine the effects of a single-dose of 1.5 mg of levonorgestrel (commonly used as emergency contraceptive) on endometrial receptivity biomarkers through the oral or vaginal route.
Plasma progesterone concentration and endometrial dating were not different. The pattern of progesterone receptors and glycodelin-A expression was not affected during the early and midsecretory phase. Some endometrial biopsies from the group in which levonorgetrel was orally administered showed areas of glandular atrophy and stromal decidualization. However, the expression of the progesterone receptor, L-selectin ligand, αvβ3 integrin, and glycodelin-A were not different between the groups.
Levonorgestrel, given as emergency contraceptive on the day of LH surge, does not disrupt either ovulation or progesterone production by the corpus luteum. The contraceptive mechanism of levonorgestrel at the time of LH surge does not include changes in the progesterone receptors or the endometrial receptivity biomarkers.
Regarding that "areas of glandular atrophy and stromal decidualization" thing, they further clarify:
In our study, the endometrial biopsies from the LNG-EC-treated subjects administered by the oral route showed areas of irregular development that were characterized by glandular atrophy and intense stromal decidualization. This morphologic change resembles the prolonged and intense effect of progestin on the endometrium. Interestingly, according to the HSCORE evaluation, neither the PR expression nor the endometrial molecular markers were affected compared with the normal endometrial tissue. Our work confirms the finding of Murray et al. (25), which demonstrates endometrial histology is not a valid method for evaluating endometrial receptivity. This finding indicates that endometrial morphology is insufficient to predict the endometrial molecular receptivity phenotype and the capability of the uterus to support embryo implantation.
In "Effects of oral and vaginal administration of levonorgestrel emergency contraception on markers of endometrial receptivity", 2010 :
METHODS Endometrial biopsies were taken from non-smoking, healthy women with proven fertility on cycle days LH + 6 to LH + 8 in control and levonorgestrel treatment cycles (each woman serving as her own control). Levonorgestrel was administered either orally (0.75 mg × 4, at 24 h intervals on LH + 1 to LH + 4; n = 8) or vaginally (a single dose of 1.5 mg on LH + 2; n = 7). Immunohistochemistry and real-time RT–PCR was performed to compare the levels of protein and mRNA for sex steroid receptors, interleukin-1β, leukaemia inhibitory factor (LIF), vascular endothelial growth factor, cyclooxygenase-2, tumour necrosis factor-α, integrin αvβ3 and mucin 1 in endometrial cells.
RESULTS Following the repeated oral treatment, the immunoreactivity of both progesterone receptor (PR)-A and PR-B declined in glandular epithelium (P = 0.03 and P = 0.02, respectively), whereas stromal immunoreactivity and mRNA expression of LIF increased compared with control (P < 0.001 and P = 0.03, respectively). However, vaginal levonorgestrel did not cause any significant endometrial changes.
CONCLUSIONS The two regimens of levonorgestrel caused either only minor or no alterations in markers of endometrial receptivity. New agents targeting the endometrial development should be explored in order to increase EC efficacy.
And in "On the mechanisms of action of short-term levonorgestrel administration in emergency contraception", 2001 :
The effects of short-term administration of levonorgestrel (LNG) at different stages of the ovarian cycle on the pituitary-ovarian axis, corpus luteum function, and endometrium were investigated. Forty-five surgically sterilized women were studied during two menstrual cycles. In the second cycle, each women received two doses of 0.75 mg LNG taken 12 h apart on day 10 of the cycle (Group A), at the time of serum luteinizing hormone (LH) surge (Group B), 48 h after positive detection of urinary LH (Group C), or late follicular phase (Group D). In both cycles, transvaginal ultrasound and serum LH were performed from the detection of urinary LH until ovulation. Serum estradiol (E2) and progesterone (P4) were measured during the complete luteal phase. In addition, an endometrial biopsy was taken at day LH + 9. Eighty percent of participants in Group A were anovulatory, the remaining (three participants) presented significant shortness of the luteal phase with notably lower luteal P4 serum concentrations. In Groups B and C, no significant differences on either cycle length or luteal P4 and E2 serum concentrations were observed between the untreated and treated cycles. Participants in Group D had normal cycle length but significantly lower luteal P4 serum concentrations. Endometrial histology was normal in all ovulatory-treated cycles. It is suggested that interference of LNG with the mechanisms initiating the LH preovulatory surge depends on the stage of follicle development. Thus, anovulation results from disrupting the normal development and/or the hormonal activity of the growing follicle only when LNG is given preovulatory. In addition, peri- and post-ovulatory administration of LNG did not impair corpus luteum function or endometrial morphology.
4. It is not clear whether levonorgestrel has any effect on pregnancy if administered shortly before ovulation, and if it does, it is not clear by what mechanism.
Any remaining controversy (such as in ; note that this is published in the journal of the Catholic Medical Association) revolves around whether levonorgestrel is effective at preventing pregnancy if administered shortly before ovulation, and if so, by what mechanism.
Levonorgestrel does not seem to prevent or delay ovulation ("follicular rupture") when administered shortly before ovulation. For example, from a study comparing levonorgestrel (LNG) and another emergency contraceptive, ulipristal acetate (UPA), "Ulipristal acetate prevents ovulation more effectively than levonorgestrel: analysis of pooled data from three randomized trials of emergency contraception regimens", 2013, :
Forty eight LNG-treated (1.5 mg) cycles, 31 LNG (1.5 mg) + meloxicam (15 mg), 34 UPA (30 mg) cycles and 50 placebo cycles were compared. Follicle rupture was delayed for at least 5 days in 14.6%, 38.7%, 58.8% and 4% of the LNG-, LNG+meloxicam-, UPA- and placebo-treated cycles, respectively. UPA was more effective than LNG and placebo in inhibiting follicular rupture (p=.0001), while LNG, when administered at this time of the cycle, was not different than placebo. The addition of meloxicam improved the efficacy of LNG in preventing follicular rupture (p=.0292 vs. LNG; p=.0001 vs. placebo; non-significant vs. UPA). UPA was effective in preventing rupture in the 5 days following treatment, even when administered at the time of the luteinizing hormone (LH) surge (UPA 79%, LNG 14% and placebo 10%). None of the treatments were effective when administered on the day of the LH peak. The median time from treatment to rupture was 6 days during the ulipristal cycles and 2 days in the placebo and LNG/LNG+meloxicam cycles (p=.0015).
Although no EC treatment is 100% effective in inhibiting follicular rupture when administered in the late follicular phase, UPA is the most effective treatment, delaying ovulation for at least 5 days in 59% of the cycles. LNG is not different from placebo in inhibiting follicular rupture at this advanced phase of the cycle. No treatment was effective in postponing rupture when administered on the day of LH peak.
If levonorgestrel does not prevent ovulation when administered at this late stage, what accounts for its effectiveness in this period? Well, first, it's not at all clear that it is effective when administered just before ovulation. For example, in another study comparing the effectiveness of levonorgestrel (which does not prevent ovulation when administered just before ovulation) and ulipristal acetate (which does), ulipristal acetate was more effective at preventing pregnancy. From "Ulipristal acetate versus levonorgestrel for emergency contraception: a randomised non-inferiority trial and meta-analysis", 2010 :
In the efficacy-evaluable population, 1696 women received emergency contraception within 72 h of sexual intercourse (ulipristal acetate, n=844; levonorgestrel, n=852). There were 15 pregnancies in the ulipristal acetate group (1·8%, 95% CI 1·0–3·0) and 22 in the levonorgestrel group (2·6%, 1·7–3·9; odds ratio [OR] 0·68, 95% CI 0·35–1·31). In 203 women who received emergency contraception between 72 h and 120 h after sexual intercourse, there were three pregnancies, all of which were in the levonorgestrel group.
It is also difficult to "account" for the effectiveness of pregnancy prevention measures (and determine whether known mechanism explain all of the effectiveness). First, this involves predicting how many pregnancies would have been expected without intervention, and the statistical models for this are sometimes unreliable. For example, in , some pregnancies occurred in women for whom "the risk of conception was theoretically nil" according to the standard model. Also, many studies measuring effectiveness of interventions under various conditions rely on women's own reports of the timing of their menstrual cycle, which is prone to error compared to other, more invasive, means .
Other studies suggest that levonorgestrel may have some effect on the ovulatory process even in cases where it is administered too late to prevent ovulation. For example, in "Pituitary–ovarian function following the standard levonorgestrel emergency contraceptive dose or a single 0.75-mg dose given on the days preceding ovulation", 2004 :
We assessed to what extent the standard dose of levonorgestrel (LNG), used for emergency contraception, or a single dose (half dose), given in the follicular phase, affects the ovulatory process during the ensuing 5-day period. Fifty-eight women were divided into three groups according to timing of treatment. Each woman contributed with three treatment cycles separated by resting cycles. All received placebo in one cycle, and standard or single dose in two other cycles, in a randomized order. The diameter of the dominant follicle determined the time of treatment. Each woman had the same diameter assigned for all her treatments. Diameters were grouped into 33 categories: 12–14, 15–17 or 18–20 mm. Follicular rupture failed to occur during the 5-day period in 44%, 50% and 36% of cycles with the standard, half dose and placebo, respectively. Ovulatory dysfunction, characterized by follicular rupture associated with absent, blunted or mistimed gonadotropin surge, occurred in 35%, 36% and 5% of standard, single dose or placebo cycles, respectively. In conclusion, LNG can disrupt the ovulatory process in 93% of cycles treated when the diameter of the dominant follicle is between 12 and 17 mm. It is highly probable that this mode of action fully accounts for the contraceptive efficacy as well as the failure rate of this method. The present data suggest that half the dose may be as effective as the standard dose.
 Novikova, N., Weisberg, E., Stanczyk, F.Z., Croxatto, H.B. and Fraser, I.S., 2007. Effectiveness of levonorgestrel emergency contraception given before or after ovulation—a pilot study. Contraception, 75(2), pp.112-118.
 Noé, G., Croxatto, H.B., Salvatierra, A.M., Reyes, V., Villarroel, C., Muñoz, C., Morales, G. and Retamales, A., 2011. Contraceptive efficacy of emergency contraception with levonorgestrel given before or after ovulation. Contraception, 84(5), pp.486-492.
 Meng, C.X., Andersson, K.L., Bentin-Ley, U., Gemzell-Danielsson, K. and Lalitkumar, P.L., 2009. Effect of levonorgestrel and mifepristone on endometrial receptivity markers in a three-dimensional human endometrial cell culture model. Fertility and sterility, 91(1), pp.256-264.
 Palomino, W.A., Kohen, P. and Devoto, L., 2010. A single midcycle dose of levonorgestrel similar to emergency contraceptive does not alter the expression of the L-selectin ligand or molecular markers of endometrial receptivity. Fertility and sterility, 94(5), pp.1589-1594.
 Meng, C.X., Marions, L., Byström, B. and Gemzell-Danielsson, K., 2010. Effects of oral and vaginal administration of levonorgestrel emergency contraception on markers of endometrial receptivity. Human Reproduction, p.deq007.
 Durand, M., del Carmen Cravioto, M., Raymond, E.G., Durán-Sánchez, O., De la Luz Cruz-Hinojosa, M., Castell-Rodrı́guez, A., Schiavon, R. and Larrea, F., 2001. On the mechanisms of action of short-term levonorgestrel administration in emergency contraception. Contraception, 64(4), pp.227-234.
 Peck, R., Rella, W., Tudela, J., Aznar, J. and Mozzanega, B., 2016. Does levonorgestrel emergency contraceptive have a post-fertilization effect? A review of its mechanism of action. The Linacre Quarterly, 83(1), pp.35-51.
 Brache, V., Cochon, L., Deniaud, M. and Croxatto, H.B., 2013. Ulipristal acetate prevents ovulation more effectively than levonorgestrel: analysis of pooled data from three randomized trials of emergency contraception regimens. Contraception, 88(5), pp.611-618.
 Glasier, A.F., Cameron, S.T., Fine, P.M., Logan, S.J., Casale, W., Van Horn, J., Sogor, L., Blithe, D.L., Scherrer, B., Mathe, H. and Jaspart, A., 2010. Ulipristal acetate versus levonorgestrel for emergency contraception: a randomised non-inferiority trial and meta-analysis. The Lancet, 375(9714), pp.555-562.
 Glasier, A., Cameron, S.T., Blithe, D., Scherrer, B., Mathe, H., Levy, D., Gainer, E. and Ulmann, A., 2011. Can we identify women at risk of pregnancy despite using emergency contraception? Data from randomized trials of ulipristal acetate and levonorgestrel. Contraception, 84(4), pp.363-367.
 Croxatto, H.B., Brache, V., Pavez, M., Cochon, L., Forcelledo, M.L., Alvarez, F., Massai, R., Faundes, A. and Salvatierra, A.M., 2004. Pituitary–ovarian function following the standard levonorgestrel emergency contraceptive dose or a single 0.75-mg dose given on the days preceding ovulation. Contraception, 70(6), pp.442-450.