Understanding Attorneys’ Plea Advice: The Role of Defendant Guilt and Trial Penalties

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Comments and Suggestions for Authors

The authors have improved the manuscript and I appreciate their attention to detail. 

My only remaining comment is to add additional citations in places for the reader to find more information.

• This sentence on p. 3 would be stronger with some citations: "Specifically, researchers have argued that variability in plea decisions may reflect individual differences (e.g., risk preferences,  implicit biases, etc.) and contextual factors (e.g., case characteristics, defendant criminal histories, etc.) that SoT does not incorporate."
• Fessinger et al (2024) in LHB is very niche (child sex abuse cases), but does also gather data from attorneys on their plea recommendations.
• The new section on Attorney Influence in Plea decisions could use more citations. Lee et al (2021) in JQ is another vignette study on attorney influence in defendant decisionmaking, Frazier & Gonzales (2021) in Psych, Crime, & Law looks at defendants' decisionmaking, including perceptions of their own guilt. Wright & Roberts (2023) in Annual Rev of Criminology is a good overview on defense attorneys in general. Shook et al (2021) analyzes juveniles, but does address the role of defense attorneys in court legitimacy views. 

Author Response

REVIEWER 1 COMMENTS:

• This sentence on p. 3 would be stronger with some citations: "Specifically, researchers have argued that variability in plea decisions may reflect individual differences (e.g., risk preferences, implicit biases, etc.) and contextual factors (e.g., case characteristics, defendant criminal histories, etc.) that SoT does not incorporate."
  • We have revised the sentence to include several citations already referenced in the surrounding text, as well as new citations to related work, to strengthen the support for this point (p. 3, lines 128–130).

• Fessinger et al (2024) in LHB is very niche (child sex abuse cases) but does also gather data from attorneys on their plea recommendations.
  • We have integrated Fessinger et al. (2024) into our discussion of the Shadow of the Trial (SoT) model in both the Introduction and the Discussion. Specifically, in the Shadow of the Trial (SoT) Model section of the Introduction, we introduced this study as recent evidence that attorneys’ plea recommendations reflect core SoT factors, noting that attorneys were more likely to recommend plea acceptance when defendants were guilty and facing harsher potential trial sentences (p. 3, lines 107–109). In the Psychological Considerations section of the Discussion, we also cited Fessinger et al. (2024) to tie our results back to similar research using SoT factors to test attorney decision-making (p. 15, lines 639–640).

• The new section on Attorney Influence in Plea decisions could use more citations. Lee et al (2021) in JQ is another vignette study on attorney influence in defendant decision-making, Frazier & Gonzales (2021) in Psych, Crime, & Law looks at defendants' decision-making, including perceptions of their own guilt. Wright & Roberts (2023) in Annual Rev of Criminology is a good overview on defense attorneys in general. Shook et al (2021) analyzes juveniles but does address the role of defense attorneys in court legitimacy views.
  • We have integrated these citations into the Attorney Influence in Plea Decisions section of the Introduction. Specifically, Lee et al. (2021) was added to reinforce experimental evidence that defendants’ plea choices are shaped by attorney recommendations (p. 2, lines 62–65); Frazier & Gonzales (2021) to show how self-perceived guilt influences how defendants interpret and weigh attorney advice (p. 2, lines 65–68); Wright & Roberts (2023) to provide a broader overview of defense attorneys’ central role in plea negotiations (p. 2, lines 73–75); and Shook et al. (2021) to highlight how attorney influence extends to perceptions of court legitimacy in juvenile contexts (p. 2, lines 78–81).

Reviewer 2 Report (Previous Reviewer 4)

Comments and Suggestions for Authors

Thank you very much for providing me with an opportunity to review the reworked manuscript. 

 

The Authors thoroughly revised the original piece, making the manuscript much more apprehensible and in-depth when it comes to the legal, ethical and psychological dimensions of the study. Generally, all of my critical comments have been addressed. Eliminating the references to anchoring, along with some other tangential issues, and focusing on Prospect Theory, definitely streamlined the paper. However, a short note (for example in the discussion section, subsections 5.3 or 5.5.) pointing out that anchoring may have played a role here and opening the door to further research would be appropriate, especially since this was the original idea of the Authors. 

 

As for other amendments made by the Authors, I am not a huge fan of the "statistical assumptions" section, although I acknowledge that another reviewer may have suggested it. Nonetheless, it could be shortened or moved to a footnote. As for minor details, the alt-texts for the figures are still odd, though this may be a fault of the submission system. 

 

Congratulations to the Authors on delivering this interesting and practically impactful study.

Author Response

REVIEWER 2 COMMENTS:

• The Authors thoroughly revised the original piece, making the manuscript much more apprehensible and in-depth when it comes to the legal, ethical and psychological dimensions of the study. Generally, all of my critical comments have been addressed. Eliminating the references to anchoring, along with some other tangential issues, and focusing on Prospect Theory, definitely streamlined the paper. However, a short note (for example in the discussion section, subsections 5.3 or 5.5.) pointing out that anchoring may have played a role here and opening the door to further research would be appropriate, especially since this was the original idea of the Authors.
  • In the Psychological Considerations section of the Discussion, we included a short note acknowledging that anchoring may have played a role. We point out that attorneys may rely on potential trial sentences as cognitive anchors that influence their risk preferences, and we leave this as an avenue for future research (p. 15, lines 646–655).

• As for other amendments made by the Authors, I am not a huge fan of the "statistical assumptions" section, although I acknowledge that another reviewer may have suggested it. Nonetheless, it could be shortened or moved to a footnote.
  • As noted, we originally included the Assumptions section in response to another reviewer’s request. We have decided to keep the assumptions in-text, but to address your and other reviewers’ comments, we significantly shortened the section and moved it to the beginning of the Results (p. 8, lines 374–384).

• As for minor details, the alt-texts for the figures are still odd, though this may be a fault of the submission system.
  • We apologize for the issue in reviewing the figures. As the figures appear normal on our end, we assume this must be an issue with the submission system. We will ensure that this is no longer a problem if the manuscript advances to publication.

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

I really appreciate the authors' work on this interesting topic. As the authors do an excellent job of stating, most criminal convictions are the result of a plea, and attorneys' recommendations are an important factor in the decision to accept a plea. The authors use Prospect Theory to examine whether the potential loss (length of the plea sentence) and the degree of the risk (factual guilt or innocence) affect attorney plea recommendations. As potential sentences following a trial conviction increased, participants were more likely to recommend the plea offer. I believe this article should be published, but I have some relatively minor suggestions the authors may want to consider. 

First, the authors have a well-researched and thought out introduction. They do a great job of explaining prospect theory and its relevance to plea decision-making. It might be helpful, though, for the authors to further explain the difference between their study and the study published by Hellgren & Kassen in 2022. There seems to be potential overlap with different results, and it might be interesting for the authors to explore in more detail why. 

The authors also lay an appropriate theoretical foundation for their population and their methods. As the authors note in their limitations section, there were differences between the CloudConnect and the snowball sample. The authors theorize that one potential reason is because most of the snowball sample participants were criminal attorneys, while the CloudConnect sample had more varied professional fields. Did the authors do any analyses to compare attorneys that identified their field as criminal law compared to those that did not? That might support their hypothesis explaining the different results. 

The results are clearly presented. I do wonder why the authors, for some of the analyses, present the results of Hypothesis 3 (the interaction effect) before then presenting the results of Hypotheses 1-2 (the main effects)? Usually I see main effects presented first, so if there is a reason for doing this perhaps the authors might want to explain it. 

The authors may also want to explore in more detail why such a small percentage of their participants would recommend the plea offer. Is it perhaps indicative of the offer being unreasonable or not worth passing up the gamble of trial? Sentences for simple drug possession can vary drastically by jurisdiction. Did the authors compare their sentences to maybe the federal sentencing guidelines or other jurisdictions outside of Nevada's criminal penalties? 

Finally, I appreciate how the authors attempted to explain their findings with respect to participants' maximum acceptable sentence they viewed. As the authors state, participants reported higher maximum acceptable sentences when the potential sentence was higher. The authors use Prospect Theory to discuss this result, but I may suggest another explanation: an anchoring bias. Perhaps participants were just falling victim to the heuristic that we tend to rely on given numbers to "anchor" our assessments? That might be interesting to discuss. 

Overall, I do recommend publication, but the authors might want to add a bit to their discussion or explanation of results to provide further clarification. 

Author Response

REVIEWER 3 COMMENTS:

• First, the authors have a well-researched and thought out introduction. They do a great job of explaining prospect theory and its relevance to plea decision-making. It might be helpful, though, for the authors to further explain the difference between their study and the study published by Hellgren & Kassen in 2022. There seems to be potential overlap with different results, and it might be interesting for the authors to explore in more detail why.
  • We agree with this assessment, and although we do not address the differences between our study and Hellgren & Kassin (2022) in the Introduction, we do so in the Implications for Attorneys’ Plea Decision-Making section of the Discussion. Specifically, we note that while Hellgren & Kassin (2022) suggest attorneys are more influenced by objective conviction probability than guilt status, our simulation was designed to establish factual guilt. Because minimal evidence was introduced and guilt status was confirmed through a flashback, attorneys may have assumed that innocent defendants were less likely to be convicted, which in turn influenced both their plea recommendations and willingness to recommend the plea (WTRP) (p. 16, lines 699–705).

• The authors also lay an appropriate theoretical foundation for their population and their methods. As the authors note in their limitations section, there were differences between the CloudConnect and the snowball sample. The authors theorize that one potential reason is because most of the snowball sample participants were criminal attorneys, while the CloudConnect sample had more varied professional fields. Did the authors do any analyses to compare attorneys that identified their field as criminal law compared to those that did not? That might support their hypothesis explaining the different results.
  • We have addressed this by incorporating additional discussion in the Limitations and Future Directions section in the Discussion. Although we could not statistically compare criminal defense attorneys’ responses to those of other attorney types due to the small number of criminal defense attorneys in our sample, we examined descriptive statistics and found that they followed the same general patterns for plea recommendations, WTRP, and maximum acceptable plea sentences, but to a lesser degree. We suspect this may be because our study did not include the negotiation dynamics that typically occur during plea bargaining, and we qualify that information with an example of an open-ended response from a criminal defense attorney (p. 17, lines 735–741).

• The results are clearly presented. I do wonder why the authors, for some of the analyses, present the results of Hypothesis 3 (the interaction effect) before then presenting the results of Hypotheses 1-2 (the main effects)? Usually, I see main effects presented first, so if there is a reason for doing this perhaps the authors might want to explain it.
  • We present the interactions prior to the main effects because a significant interaction could influence the interpretation of main effects. Particularly, interactions are conditional effects, whereas main effects are averaged effects. If there exists a significant interaction, the main effects can be misleading when interpreted alone. To address your comment, we now note in the Results section that we analyzed interactions first for this reason (p. 9, lines 409–411).

• The authors may also want to explore in more detail why such a small percentage of their participants would recommend the plea offer. Is it perhaps indicative of the offer being unreasonable or not worth passing up the gamble of trial? Sentences for simple drug possession can vary drastically by jurisdiction. Did the authors compare their sentences to maybe the federal sentencing guidelines or other jurisdictions outside of Nevada's criminal penalties?
  • We have addressed your concerns in two ways. First, we received concern regarding the percentage of participants recommending the plea offer in the last round of revisions, and we added more detailed descriptive statistics in the Plea Recommendations section of the Results to address this (p. 9, lines 386–393). Overall, 41% of our sample recommended plea acceptance. Even in the high-sentence conditions that were designed to simulate an unreasonable offer (i.e., overcharging), 30% of our sample still recommended plea acceptance. Although this percentage is not reflective of the rate of plea acceptances in real practice, we do not interpret it as a floor effect given the lack of negotiation in our study design. Second, while we were not able to compare our sentences to federal sentencing guidelines or those of other jurisdictions outside Nevada, we accept this as a limitation of our results and have added language to the Limitations and Future Directions section of the Discussion to reflect this point (p. 18, lines 776–777).

• Finally, I appreciate how the authors attempted to explain their findings with respect to participants' maximum acceptable sentence they viewed. As the authors state, participants reported higher maximum acceptable sentences when the potential sentence was higher. The authors use Prospect Theory to discuss this result, but I may suggest another explanation: an anchoring bias. Perhaps participants were just falling victim to the heuristic that we tend to rely on given numbers to "anchor" our assessments? That might be interesting to discuss.

In our prior revision, we removed discussion of anchoring because other reviewers raised concerns about using it to explain our results. However, to address your and other reviewers’ comments, we added a short note in the Psychological Considerations section of the Discussion acknowledging that anchoring may have played a role. Specifically, we point out that attorneys may rely on potential trial sentences as cognitive anchors that shape their risk preferences, and we leave this as an avenue for future research (p. 15, lines 646–655).

Reviewer 4 Report (New Reviewer)

Comments and Suggestions for Authors

Thank you for the opportunity to review “Understanding Attorneys’ Plea Advice: The Role of Defendant 2 Guilt and Trial Penalties”. The authors present a one-study investigation into how potential trial sentence and defendant guilt status impact attorney’s plea recommendations and maximum acceptable plea sentences. I commend the authors on their well-articulated, clear, logical structure of the manuscript and believe their research question adds to the literature. I also particularly believe that the large sample of practicing attorneys is quite a strength of their work. I have just a few minor points/suggestions that I believe would improve the manuscript.

1. On page 5 can the authors specific the results of the pilot study in terms of participants’ classifications of reasonable, moderate, or excessive punishment?
2. Can the authors say a bit more about the prescreening questions for Cloud Research participants to determine their practicing attorney status? Given that some results did vary across sampling type, I think a bit more detail on what the prescreening questions mentioned on page 6 entailed would be beneficial.
3. I would suggest moving the “Assumptions” section from the end of the results to the beginning where you discuss descriptives, etc., on page 8. It felt awkward to read after already reading the findings from the models. 
4. While I know the authors’ primary focus was examining Prospect Theory, I believe at least a mention of how the results relate to Shadow of a Trial theory in the discussion would be beneficial. Given that the authors spend a full subsection delineating the theory in their introduction, I was looking for some tieback in the discussion.

Author Response

REVIEWER 4 COMMENTS:

• On page 5 can the authors specific the results of the pilot study in terms of participants’ classifications of reasonable, moderate, or excessive punishment?
  • We have incorporated the pilot study results on by specifying how participants classified each trial sentence condition. Specifically, participants rated the 2-year sentence as 75% reasonable and 25% moderate, the 4-year sentence as 80% moderate and 20% excessive, and the 20-year sentence as 100% excessive (p. 6, lines 242–243).

• Can the authors say a bit more about the prescreening questions for Cloud Research participants to determine their practicing attorney status? Given that some results did vary across sampling type, I think a bit more detail on what the prescreening questions mentioned on page 6 entailed would be beneficial.
  • We have addressed this by expanding the description of our prescreening procedures. In the CloudResearch (CR) Connect platform, we verified that participants worked in the legal industry and held a law degree. In the survey, we confirmed that they were practicing attorneys and asked what type of law they practice (p. 6, lines 259–263).

• I would suggest moving the “Assumptions” section from the end of the results to the beginning where you discuss descriptives, etc., on page 8. It felt awkward to read after already reading the findings from the models.
  • To address your and other reviewers’ comments, we have shortened the assumptions and moved it to the beginning of the Results (p. 8, lines 374–384).

• While I know the authors’ primary focus was examining Prospect Theory, I believe at least a mention of how the results relate to Shadow of a Trial theory in the discussion would be beneficial. Given that the authors spend a full subsection delineating the theory in their introduction, I was looking for some tieback in the discussion.
  • In the Psychological Considerations section of the Discussion, we tied our findings back to the Shadow of the Trial (SoT) model. We note that potential trial sentence and defendant guilt status independently shaped attorneys’ plea recommendations and willingness to recommend the plea (WTRP), and that their interaction influenced maximum acceptable plea sentences. This reinforces the predictive value of SoT factors while adding nuance about how they may interact to shape attorney risk preferences (p. 15, lines 634–659).

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Behavioral Sciences – 3694854: “Reference-Dependent Anchoring in Attorneys’ Plea Decision-Making”

This paper uses a unique sample of practicing attorneys to examine how guilt status and potential trial sentence affect whether they believe they would recommend that their clients accept a plea. The authors find that attorneys were more likely to recommend acceptance for guilty clients and in situations with lower potential maximum sentences. The paper is well written and I enjoyed reading it. I also appreciate the authors’ data collection, as well as their interesting and policy-relevant research questions. I believe the authors could expand on some additional relevant studies, acknowledge a few more limitations, and tie their results back into extant literature more.

1. The authors focus their hook on p. 1 on false pleas, which are obviously very important. I think they could spend more time here discussing the vignette studies (that are cited to later on) showing how influential attorneys are on clients’ plea choices. Criminal defendants are really at a loss for information and their attorney is often the sole source of advice in this scenario. Suiter and Metcalfe (2024) is a recent study on defense attorneys that could also be included.
2. The authors should include Edkins (2011, Law and Human Behavior) That is one of the few other studies that studies attorneys as opposed to defendants or hypothetical defendants. They may also consider whether any of the Shadow of the Trial research by Bushway and colleagues would be relevant here, as they also surveyed courtroom actors.
3. Tor et al (2010) in JELS might also be another good one to explore; I think their discussion of fairness is an interesting corollary to prospect theory.
4. I appreciate the use of a novel vignette and the manipulations (guilt, max sentence at trial) make sense. I do have a few thoughts about the vignette: 1) I wonder if all the penalties (2, 10, 20 years) all seem realistic to attorneys? Some of those seem very very high, even in states that are “tough on crime.” The authors get at this somewhat with their discussion on p. 11, lines 426-429, and I think it could use more attention; 2) The vignette is set up so that the prosecutor says they will seek the highest penalty if the case goes to trial, but judges determine sentences, not prosecutors. I think the authors should at least acknowledge this; 3) I do also wonder how realistic it is for clients to tell their attorneys “I am guilty” or “I am not guilty.” Again, this is not a death knell for the paper, but I would suggest adding some mention in the limitations that it’s hard to say if the conversations in the vignette are realistic for practicing attorneys.
5. I have a few questions about the data collection: 1) did any participants make it through steps one and two (p. 6) and then not finish the remainder of the Qualtrics questions? If so, how many?; 2) How many questions were in the survey total?; 3) Were any other vignettes/simulations included? I was slightly confused by some of the language on p. 6.
6. It might be a journal or disciplinary difference, but I was surprised not to see any information on the sample demographics or a descriptive table. Could the authors add any of that into the text or a table?
7. I think the authors could also acknowledge in the limitations that this was only for a drug case and guilt was clear. Other case types and less clear culpability may result in different findings.
8. After the authors include some of the suggested papers above, they could elaborate more in the discussion about how their results differ from extant research – especially Edkins (2011). That paper provides an interesting opportunity for comparison of one of its dependent variables (the best sentence they think they could get for their client) to one of the present study’s outcome variables (maximum acceptable plea sentence).

Reviewer 2 Report

Comments and Suggestions for Authors

See attached document.

Comments for author File: Comments.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors consider the effect of guilt status and sentence length on attorney plea decisions. In an experiment on attorneys, they tested the effect of these factors on plea acceptance, willingness to recommend a plea (WTRP), and maximum acceptable plea (MAP). Plea acceptance and WTRP were higher for guilty defendants and for low and moderate sentences, with no interaction. MAP increased with sentence length which interacted with guilt, but guilt itself was not significant. The authors interpret the results through the lens of anchoring, reference dependence, and loss aversion.

The manuscript is well written, easy to follow, and a good length. I am not directly familiar with this literature, so I cannot comment on the novelty of the research question or the literature review with confidence, but the question seems of practical importance. I appreciate the power analysis.

My major concern revolves around the interpretation of the results in terms of heuristics (and related concepts). Although a potentially good motivation, this conclusion seems premature, especially given the experimental design (see below). Heuristics are often (roughly, and not universally), defined as simplifying decision making strategies that deviate from rationality or optimality. It is unclear whether that holds here. There is no discussion or theory development of why these are heuristics, what is being simplified, whether simplification is needed, etc. Why anchoring on sentence length? That is especially problematic as sentence length is confounded with plea length. Why reference dependence and loss aversion? Can't guilt directly affect this factor via conviction probability combined with sentence length? The authors dismiss this in the discussion, but perhaps too quickly. If you assume conviction probabilities of .3 and .15 (guilt/innocence) and simply multiply by sentence length you get exactly figure 5. You can them compare these values to the plea lengths to get acceptance probabilities. (Adding noise to this simple model also predicts that low and moderate sentences will be accepted more often, although I think it predicts that low is accepted more than moderate, which doesn't match the data.) Using this very simple model shows that plea length is much higher than expected sentence length for the max sentence relative to low and medium, which would lower the probability of accepting a plea in this condition, exactly as was seen in the data. This difference was higher for innocent defendants, again, as was seen in the data. I'm not saying that model is right (in fact, I'm sure it's not), but the point is that, before invoking heuristics, it is necessary to have a more complete picture of how the factor "should" be combined. A relationship to a factor does not imply use of a heuristic. Even if we accept heuristics are being used, I would like to see a deeper analysis into how they are being used, e.g., a function from the factors to the result.

Were the regression assumptions met? In particular, for WTRP and MAP. The non-significant effect of guilt seems odd (F5). Could that be because a lot of the data are near 0 and the DV is bounded? I'm guessing normality and HOV are violated. 

Minor: 

Please note how guilt was coded (I'm assuming 0=guilty).

To better allow the reader to evaluate these variables, please provide continuous graphs of WTRP and MAP. 

I understand the interaction was not significant, but, for acceptance rates and WTRP, please show all combinations of conditions (as in Fig 5). 

I am on the fence regarding attorney knowledge of guilt. I would have liked to have seen a condition in which guilt is not known.

Reviewer 4 Report

Comments and Suggestions for Authors

The paper presents an experimental study on anchoring and guilt status in plea-bargaining decision making. The study has notable practical implications, as it suggests that anchored propositions may exert a substantial influence on attorneys’ decision-making in the context of plea bargaining, thereby contributing meaningfully to ongoing discussions about the boundaries of plea practices in criminal proceedings. Moreover, it has a potential to contribute to the psychological knowledge about plea decision-making processes, the anchoring effect, and the prospect theory. Even more importantly, the Authors’ use of a sample composed of professional attorneys and the employment of simulated trial settings (rather than relying solely on written vignettes) enhance the ecological validity of the findings and are particularly praiseworthy.

 

That being said, the current version of the manuscript presents some limitations, particularly in relation to the theoretical framework, which would benefit from further development. While it is clear that this is primarily an experimental contribution rather than a theoretical or doctrinal analysis in a law review, the non-empirical sections of the paper could be strengthened to enhance the overall impact of the work. Additionally, there are some more minor concerns regarding aspects of the study design that warrant attention.

 

First, the psychological mechanisms underlying the anchoring effect – which constitutes the central focus of the experiment – are not sufficiently discussed. The original “anchoring and adjustment” model proposed by Tversky and Kahneman is an important foundation, but it is not the only theoretical account of this phenomenon. For instance, recent reviews (e.g., Bahník, Š., Mussweiler, T., & Strack, F., 2022, Anchoring Effect, in Cognitive Illusions, Routledge) highlight multiple explanatory frameworks, including both semantic (meaning-based) and more “shallow” numerical aspects of the phenomenon. In addition, at least three relevant meta-analyses explore anchoring in domains closely aligned with the current study such as negotiations, legal decision-making, and willingness-to-X paradigms.

 

Second, the discussion of the psychological dimensions of plea bargaining could benefit from further elaboration, as well as a more in-depth look at its ethical, strategic, and legal dimensions. While the reference list includes several pertinent studies, some of these are not cited or integrated into the main text. For instance, recent work by Rebecca Helm (The Psychology of Guilty Plea Decisions; Plea-Based Sentence Reductions: Legal Assumptions and Empirical Realities) could offer valuable insights. Additionally, a particularly relevant and very recent study that appears to be missing is the paper by Valerie Reyna and colleagues, Framing Biases in Plea Bargaining Decisions in Those With and Without Criminal Involvement: Tests of Theoretical Assumptions (Journal of Behavioral Decision Making, 2025), which directly engages with psychological mechanisms in plea contexts. Incorporating these could substantially enhance the theoretical and empirical grounding of the manuscript.

 

The only larger behavioral theory explicitly mentioned by the Authors is Prospect Theory, which, while relevant, does not appear to fully account for the findings presented. For instance, Fuzzy Trace Theory may offer alternative or complementary explanatory mechanisms that could help contextualize some of the observations. These are just general remarks – in the following sections, I will aim to provide more specific comments regarding points where the theoretical parts might be expanded or where the current limitations are particularly salient.

 

The Authors cite Prospect Theory, and reference dependence in particular, as the main  theoretical foundation for the experimental design. Reference dependence is related to framing a given outcome as gain or loss, even if the expected value of the choice is the same. It would be helpful for the authors to clarify what the reference frame is exactly in the given experimental settings. For example, is the reference point in the "Guilty" condition something akin to “I am going to jail,” making the plea offer appear as a relative gain, whereas in the "Innocent" condition it is “I should be free,” thereby framing the offer as a loss? If this interpretation is correct, making it more explicit would be worthwhile.

Furthermore, in the context of anchoring, the lack of relevant numerical reference point is the key issue (see for example scale distortion approach). This suggests that anchoring and reference dependence may operate through distinct mechanisms – anchoring being primarily numerical or comparative, while reference dependence is more about mode of presentation. As such, anchoring should be distinguished from the framing as gain/loss; and is not clear how the reference dependence would work when it comes to numerical, i.e., estimation tasks. In other words, there may be two completely different reference points for the decisions on guilt and sentencing (although the two may potentially interact, cf. the discussions on so-called severity-leniency hypothesis). This is an important thing when it comes to Hypothesis 3 (l.143-150).

At several points in the manuscript, particularly around l. 113-115, the Authors seems to suggest that there exists an objective, normative standard against which plea recommendations or decisions can be evaluated. This is a problematic assumption, given the inherently strategic nature of plea bargaining and the normative-evaluative aspects of legal reasoning and interpretation. In this regard, the analogy to pain-and-suffering damages may be useful: although no single “correct” amount of money to be given to defendants exists, anchoring effects are nonetheless recognized as a potential source of undue influence. A very relevant discussion can be found in Reyna et al. (2025, p. 11), which the authors may wish to engage with more directly. These concerns also resurface in the Discussion section, particularly in l. 463-465 (on the margin, this passage might be more appropriately placed in the following section) and in the discussion of “ethical considerations and objective evaluations” (l. 469-477). Evaluations cannot be objective by definition. It would be helpful for the Authors to clarify what is meant by “objective factors” in this context. One could argue, for example, that the sentencing plea proposition is an absolutely relevant factor that should be taken into consideration. Maybe incorporating some findings from the open-ended responses provided by participants would be a good idea (especially since the reason-giving component of the study is very interesting). Finally, regarding the statement on l. 545–546 about the bias being “suboptimal for all defendants”: it may be more accurate to frame the concern in terms of fairness or procedural justice, rather than global suboptimality. Such a reframing could more precisely reflect the normative concerns at stake.

The fragment on l. 95–97 would benefit from appropriate citations to support the statements made. Additionally, in several places (l. 116, 392, and 539-540), the Authors refer to multiple “heuristics,” while the text primarily discusses anchoring (and possibly some framing via reference dependence). It may be more accurate to refer to these as “biases” or, even more neutrally, as “cognitive effects.” The term “heuristics” typically implies the use of some kind of simple rule-based, problem-solving strategies (even if they are implicit), which may or may not be operative here.

 

In the description of Hypothesis 3 (l. 147–150), it is suggested that greater proposed sentences result in greater perceived gains from plea offers. While this holds in nominal terms (e.g., a 10-year discount versus a 2-year discount), it may not hold in relative or psychological terms, due to the  – an aspect of Prospect Theory that is not currently given much attention. For instance, from the perspective of a defendant, both 10 and 20 years may be perceived similarly as “a very long time,” whereas differences between 2 and 1 year may be more graspable. In terms of FTT, both options would have the same gist judgments, despite different verbatim judgements. The diminishing sensitivity is also an important issue as anchor magnitude, as some studies show that the effect is weakened when extreme anchors appear (see the meta-analyses). The results (graph on p. 306; see l. 426-430) also seem to point in that direction – it would be helpful for the Authors to reflect on this in the Discussion. Additionally, how were the numbers that served as anchors (l.206-208) chosen? Were these based on a pre-test, actual sentencing practice, or selected in some other way? Clarification on this point would help readers assess the plausibility of the experimental manipulation. The Authors (l. 428) describe the high anchor explicitly as unreasonable, and indeed it appears considerably higher than what would be expected from a linear increase across conditions. Was that intentional and clear to the participants (cf. Section “Implausibly High Anchors” in Cardenas 2023)?

When it comes to issues related to the sample, it is appreciated that the Authors acknowledge some potential limitations (e.g., l. 518-526). Recruiting more specialized attorneys, particularly those with significant experience in criminal or drug-related cases, is understandably challenging. It would be however helpful to know whether any data were collected on participants’ professional experience (including drug cases). Throughout the manuscript, the term “attorney-participants” is used. This phrasing might inadvertently suggest the inclusion of non-attorney participants in the study. Since the sample appears to be composed exclusively of attorneys, referring to them simply as “participants” may be more precise and less ambiguous in this context. Of course, a comparative study including laypeople would be highly valuable, but in the current study design that distinction does not appear to be relevant. Regarding the passage on l. 496-500, a comparison of attorneys’ susceptibility to biases with judges or and legal professionals would be in order. For instance, prior studies on judges and prosecutors (e.g., the works of Jeffrey Rachlinski and Birte Englich, respectively) have shown that legal professionals are far from being immune to anchoring effects, even in contexts that they deal with on a daily basis. This would also alleviate the problem described on l. 518-526 – do the Authors, in the light of general susceptibility of legal professionals (and experts in their fields in general, recall famous study by Northcraft and Neale 1987) to anchoring, expect that the results would be different in case of specialized criminal attorneys? If so, why would that be?

Regarding the study design, some clarification would be helpful on the relationship between the two dependent variables, namely the binary plea acceptance recommendations and WTRP measure. At present, it is not entirely clear how distinct these two measures are conceptually and empirically. Might they reflect the same underlying attitude toward the plea offer in different formats? What is their relation to each other? Additionally, were the participants asked to render their maximum sentence in months or years (cf. l. 355)? When it comes to anchoring, the format may be quite important (cf., e.g., McAuliff & Bornstein, 2010, All Anchors Are Not Created Equal, Law and Human Behavior).

A brief clarification would be welcome regarding the hypothesized lack of interaction (l. 288-290, 329-330). Specifically, is this expectation consistent with reference dependence as outlined in PT? If so, it would be helpful for the Authors to explain why reference dependence would predict main effects but no interaction between guilt status and anchor magnitude. 

 

Some parts of the Discussion section would benefit from more depth and clarification. In particular, the structure might be streamlined by incorporating the content of Section 5.3 into Sections 5.1 and 5.2. The specific fragments that could benefit from refining include:

• l. 395-400 – it would be helpful to clarify the precise nature of the anchoring observed in this study. Did anchoring influence only the numerical estimation, or did it also affect more substantive aspects of plea judgment? Some studies suggest (e.g., Feldman, Schurr, & Teichman, 2016, Anchoring Legal Standards) that numerical anchors can sometimes influence interpretation of a legal standard itself.
• l. 432-438 – how do the observed effects relate to PT and risk attitudes?
• l. 435-438 – the reference to “thought processes” could be further clarified. A brief elaboration on the psychological or strategic drivers of the results would be valuable here
• l. 451-453 – does it, according to the Authors, mean that anchoring is stronger in the Guilty conditions? It would be helpful to address this more directly. Do the Authors have a theoretical or empirical explanation for why that might be the case?
• l.458-460 – how was the anchoring reference-dependent? Were some anchors seen as gains, and some as losses?

 

The argument made on l. 504-516, that certain plea discounts may be very problematic, contrary to two landmark precedents, is a particularly important takeaway message. Given its potential public significance for legal scholars and practitioners alike, this point might be worth reiterating explicitly in the Conclusions section. Doing so would give the paper a clearer normative takeaway and help avoid vague statements about countering biases.

 

Finally, although the paper is written very diligently, please check the manuscript for some typos etc. (e.g., “Prospect Thoery” in l. 45, or weird automatically generated alt texts for figures).

Reviewer 5 Report

Comments and Suggestions for Authors

The researcher sought to examine whether loss aversion and the anchoring-and-adjustment heuristic influence attorneys’ plea recommendations. Unfortunately, I do not believe that the authors have adequately examined these theories in the current study, and as a consequence the conclusions may not be valid. I have a few overarching areas of concern, and then some smaller specific issues that the authors should consider addressing.

1. Based on the hypotheses, it seems as though the authors are testing risk-seeking and aversion, as opposed to losses and gains as described by prospect theory.

 

2. I would like the authors to explain in greater detail why they think that it is problematic for attorneys to be using anchoring. Information about a potential trial sentence is very relevant and should factor into an attorney’s recommendation. They should be rationally weighing this against evidence, probability of conviction etc. (not mentioned in this article). Thus in my opinion the PTS does not function as heuristic but as an important data point in decision making, and need a stronger argument from the authors to be convinced otherwise.

 

3. The stimulus itself is somewhat problematic. The individual is charged with “drug possession” which is vague – in what degree; and is 20 years even in the realm of possibility as a sentence for possession in what has to be a small amount if a person is buying it at a party. I do think it is important to have ecological validity especially when the participants are practicing attorneys who would know what likely penalties are in these sorts of cases. It’s possible this is why they didn’t see some of their expected results. There is a significant difference between a drop from years to 1 year, as compared to 20 years to 10 years. I would really encourage the authors to look at some of the new work by Zottoli and Helm that explores this – I am surprised that their work was not cited here actually, and that is definitely a miss on the author’s part.

 

4. No interaction between guilt status and PTS was found on acceptance/rejection of plea (there was an effect on the arbitrary max sentence acceptable variable but given that is not a legally relevant outcome I don’t think the findings there should be given much or any weight), so I would argue that there was no support for loss/gain as discussed in the discussion section.

 

5. I am concerned that the authors are over-stating the implications of this work to actual practice, especially given my concerns with the validity of the PTS manipulation and I fear they may have lost some sensitivity in their analysis because of it. Additionally, they state that attorneys shouldn’t use heuristics but should base decisions on evidence, but they didn’t provide any evidence disclosure/manipulation to the attorneys so they do not know if attorneys in fact do use evidence strength in making recommendations (indeed research shows that they do if it is available to them at the time of plea bargaining). The authors conclude that attorney’s think similarly to defendants and this is just not a sophisticated enough study to base that conclusion on. It is experimentally fine, but the manipulations are not sensitive enough to make this claim. Thus, based on the results presented it seems as though the authors are over-reaching in terms of the implications of this work.

Other issues

• Line 65 should say expected value at trial not of trial
• Really can’t refer to attorney’s making recommendations based on the potential trial sentence “biased.” These recommendations are based on what information they have at hand, and what the potential outcome might be at trial given their client’s guilt status.
• I don’t believe the existing research shows directly (as is stated on page 3) that atty recommendations can contribute to false guilty pleas. Indeed, at least one of the articles cited to support this notion does not discuss the issue at all in this context, so I am doubtful of the validity of the other cites related to this. I would encourage to authors to re-check.
• The authors state in their discussion, “Potential trial sentence also had a significant impact on attorney-participants’ maximum acceptable plea sentences, even in the high potential trial sentence (overcharging) condition. This finding clearly demonstrates that attorneys are also vulnerable to the effects of anchoring-and- adjustment and lend support to recommendations that address sentence disparities between trial and plea outcomes could help increase fairness in the process.” – The discussion is the first time overcharging is mentioned and used as a justification for the 20 year condition. The findings don’t support, at least not as strongly as the authors state here, but in my opinion not at all, that the anchoring is at play here as I explain above.