*212OPINION OF THE COURT
STAPLETON, Circuit Judge:Connie Habecker and John Michael Ha-becker, decedent’s wife and minor child, claim that defendants are strictly liable for John Habecker’s death under the doctrine of “crashworthiness” because a forklift without an operator restraint system is a defective product. Appealing for the second time from a jury verdict for defendants, the Habeckers raise two major claims.1 First, they argue that the district court erred in admitting evidence of what was known in the industry when the forklift was manufactured about the desirability of operator restraint systems, evidence that they claim is irrelevant in a strict liability case. Second, they claim that the district court erroneously believed that it could not permit new legal theories and evidence on retrial. Because we conclude that evidence of what was known at the time of manufacture about the desirability of a safety feature is irrelevant in a crash-worthiness case and that a substantial possibility exists that the inadmissible evidence influenced the jury, we will reverse the district court’s judgment and remand for a new trial. On remand, the district court may exercise its discretion as to whether the Habeckers should be allowed to present new theories not barred by this court’s mandate or its own prior orders that were not appealed.
I.
John Habecker was killed while attempting to back a forklift down a ramp. The forklift went off the side of the ramp; Habecker was thrown out and crushed beneath the forklift. The forklift, manufactured by Clark Equipment Company (“Clark”) and leased by Forklifts, Inc., (“Forklifts”) was not equipped with operator restraints, such as a seat belt or harness. The accident happened during the course of Habecker’s employment as a civilian employee of the New Cumberland Army Depot.
Although this case originally included implied warranty and negligence claims, those claims were eliminated from the case before the first trial. Once the case was limited to strict liability, Forklifts appeared to be indemnified by Clark for any potential liability; therefore, in a stipulation signed by all the parties, Forklifts dismissed with prejudice its third-party claims against the Department of the Army and the United States.
After the first trial, the district court granted a directed verdict for the manufacturer of the ramp and a directed verdict for Clark and Forklifts on the issue of whether the forklift’s throttle was defective. The question of whether Clark and Forklifts were strictly liable for the absence of operator restraints in the forklift went to the jury, which ruled for defendants. On appeal, this court concluded that the district court had violated Rule 702 of the Federal Rules of Evidence by refusing to permit one of the Habeckers’ expert witnesses to testify and that this exclusion could well have affected plaintiffs’ substantive rights. 893 F.2d 49, 53 (3d Cir.1990). Therefore, this court upheld the directed verdicts, but remanded the case “for a new trial against defendants Clark and Forklifts on the operator restraints issue.” Id. at 54.
Throughout the first trial, the Habeckers argued only that the forklift was defective at the time it was manufactured — making the liability of Clark and Forklifts coterminous. Before the second trial, however, the Habeckers stated for the first time that they wished to pursue an additional theory of strict liability against Forklifts based on the hypothesis that Forklifts had placed the forklift in the stream of commerce in a defective condition when it leased the forklift to the Army in 1983 and 1984. The district court refused to allow that theory or any other theory that had not been asserted at the first trial, apparently concluding that our mandate deprived it of discretion to permit the assertion of new theories of liability. The district court also refused to allow any evidence or witnesses *213not listed for the first trial. The Habeck-ers then agreed to dismiss Forklifts, without prejudice to their right to argue on appeal that they were entitled to pursue liability arising from the leases. The trial was held from June 18, 1990 to June 25, 1990, and a jury verdict was returned in favor of defendant. Final judgment was entered in favor of defendants, and the Habeckers filed this timely appeal.
II.
In a motion in limine and throughout the trial, the Habeckers consistently objected to evidence concerning what the industry knew about the efficacy of operator restraint systems in 1977. In particular, they objected unsuccessfully to evidence offered by Clark tending to show that the industry had been unable to determine whether operator restraint systems reduced the risk of serious injury to the operator and that it was only later, after the development of more sophisticated computer modeling, that the industry decided it was desirable to equip forklifts with such systems. As of 1977, Clark’s evidence suggested, a manufacturer simply could not have known whether operator restraint systems created more risks than they eliminated.
The Habeckers claim that this evidence was inadmissible and, combined with the defendant’s closing argument, improperly focused the jury’s attention on Clark’s conduct, rather than on the forklift itself. Because Clark conceded that operator restraint systems were feasible at the time the forklift was manufactured, the Habeck-ers argue that the only question for the jury was whether-seen in light of all the evidence available today-a forklift without an operator restraint system is “defective.” While they acknowledge that evidence of whether a restraint system results in a net reduction in the risk of serious injury is relevant to that issue, the Habeck-ers insist that evidence of whether one could have known of that net reduction in 1977, including evidence of computer modeling capability at that time, is irrelevant to that issue. For the reasons that follow, we agree.
A.
Section 402A of the Restatement (Second) of Torts, adopted in Pennsylvania, states in relevant part,
(1) One who sells any product in a defective condition unreasonably dangerous to the user or consumer or to his property is subject to liability for physical harm thereby caused....
(2) The rule stated in Subsection (1) applies although
(a) the seller has exercised all possible care in the preparation and sale of his product....
Pennsylvania law recognizes that a product may be “defective” not only because of an error in manufacturing, but also because of its design:
the jury may find a defect where the product left the supplier’s control lacking any element necessary to make it safe for its intended use or possessing any feature that renders it unsafe for the intended use.
Azzarello v. Black Brothers Co., 480 Pa. 547, 391 A.2d 1020, 1027 (1978).2 Under the doctrine of crashworthiness, the manufacturer’s liability for producing defectively designed products includes liability for failing to provide safety features and liability for providing inadequate safety features. As this court has noted,
crashworthiness ... is a variation of strict liability theory.... [It] extends the manufacturer’s liability to situations in which the defect did not cause the accident or initial impact, but rather increased the severity of the injury over that which would have occurred absent the defective design.
*214Barris v. Bob’s Drag Chutes & Equipment, 685 F.2d 94, 99 (3d Cir.1982). Thus, crashworthiness imposes liability on the manufacturer for enhancing or failing to minimize injuries suffered in an accident, rather than for causing the accident.
This court first addressed crashworthiness in Huddell v. Levin, 537 F.2d 726 (3d Cir.1976), a case decided under New Jersey law. In that case, Dr. Huddell’s car stalled when he ran out of gas and was struck from behind by another vehicle “at a considerable rate of speed.” The impact caused Dr. Huddell’s head to hit his head restraint at about 10 m.p.h. Dr. Huddell died one day later of brain damage. His wife sued the driver of the other vehicle, but also brought suit against the manufacturer on the theory that the head restraint was defectively designed.
This court began with the legal question of whether the manufacturer owed a duty to Mr. Huddell.
We take it as beyond peradventure that an automobile manufacturer today has some legal obligation to design and produce a reasonably crashworthy vehicle. The manufacturer is not required to design against bizarre accidents; the manufacturer is not required to produce an accident-proof vehicle. But the manufacturer is required to take reasonable steps — within the limitations of cost, technology, and marketability — to design and produce a vehicle that will minimize the unavoidable danger_ [T]he manufacturer must consider accidents as among the “intended” uses of its product.
537 F.2d at 735.
Having found both that the manufacturer owed Dr. Huddell this duty and that Mrs. Huddell had presented evidence to suggest the head restraint was defective, this court held that it was for the jury to determine whether the restraint was defective and whether the defect caused or enhanced Dr. Huddell’s injuries. Huddell established the following burden of proof for plaintiffs in crashworthiness cases.
Unlike orthodox products liability or negligence litigation, crashworthy or second collision cases impugning the design of an automobile require a highly refined and almost invariably difficult presentation of proof in three aspects. First, in establishing that the design in question was defective, the plaintiff must offer proof of an alternative, safer design, practicable under the circumstances.... Second, the plaintiff must offer proof of what injuries, if any, would have resulted had the alternative, safer design been used....
Third, as a corollary to the second aspect of proof, the plaintiff must offer some method of establishing the extent of enhanced injuries attributable to the defective design.
Huddell v. Levin, 537 F.2d 726, 737-38 (3d Cir.1976).
Although the Pennsylvania Supreme Court has not had occasion to adopt or reject the crashworthiness doctrine, this court has held that “[t]he crashworthiness doctrine as articulated in Huddell v. Levin is applicable to cases governed by Pennsylvania law.” Roe v. Deere and Co., 855 F.2d 151, 153 n. 2 (3d Cir.1988); see also Jeng v. Witters, 452 F.Supp. 1349 (M.D.Pa. 1978), aff'd mem. 591 F.2d 1334 (3d Cir. 1979). In Roe, a tractor rollover case, we concluded that “[c]rashworthiness is merely a subset of strict liability.” 855 F.2d at 154. Because plaintiff had pled strict liability and had presented evidence that the tractor was uncrashworthy, we held that the district court had erred in refusing to charge the jury on the crashworthiness doctrine.
B.
Having predicted that the Pennsylvania Supreme Court would adopt the crashwor-thiness doctrine, we are required by this case to predict what definition that court would give that doctrine and, specifically, what evidence it would find relevant in determining liability. The Habeckers argue that it was error for the district court to admit evidence concerning computer modeling capabilities in 1977, because that evidence shifted the focus from the defectiveness of the product to the knowledge *215and conduct of the manufacturer. Clark responds that the Supreme Court of Pennsylvania would admit such evidence because excluding it would impose strict liability bounded only by the ability of “the fruitful imagination of expert witnesses” to describe safer alternative designs. In essence, Clark argues that a product is not “defective” because of its design if no one could know at the time of manufacture that an alternative design would be safer.
It is important to distinguish evidence of what safety features were feasible at the time a product was designed, from evidence of what safety features were known to be desirable at that time. Evidence of the first must be admissible in a crashworthiness case. Liability is imposed on a manufacturer in a crashworthiness case for a design defect because an alternative, feasible, safer design would have lessened or eliminated the injury plaintiff suffered. If no such alternative feasible design existed when the product was manufactured, then the design cannot be said to be “defective”, even if more recent technology has rendered a safer design feasible. Therefore, the fact finder can only determine whether a particular design was defective after hearing evidence about what designs were feasible at the time the product was manufactured and whether they were in fact safer. Roe, 855 F.2d at 155 (exclusion of testimony on effectiveness of safety device was error); see also Huddell, 537 F.2d at 737; Craigie v. General Motors Corporation, 740 F.Supp. 353, 357 n. 8 (E.D.Pa.1990).
While evidence of a safety feature’s feasibility is relevant to whether the product was defective under Pennsylvania law, evidence of what was known about the desirability of a safety feature is not. One commentator has explained that exclusion of evidence of foreseeability is, indeed, the distinction between strict liability for design defects and liability for negligence in designing products. Strict liability for design defect is distinguishable from liability for negligence only if the actual risks and benefits of the design are evaluated in a strict liability case without any reference to the foreseeable risks and benefits:
The fact that a risk or hazard related to the use of a product was not discoverable under existing technology and in the exercise of utmost care ... [is] irrelevant if the product itself and not defendant’s conduct is being evaluated.
Prosser & Keeton, Torts § 99 at 700 (5th ed. 1984 & Supp.1988); see also Note, Perpetuating Negligence Principles in Strict Products Liability: The Use of State of the Art Concepts in Design Cases, 36 Syracuse L.Rev. 759, 804 (1985) (“Strict products liability differs from negligence in design cases in that a manufacturer is assumed to have knowledge of all hazards [however unforeseeable] posed by its product.”).
Excluding evidence of what was or was not known about the desirability of operator restraint systems in 1977 is consistent with Pennsylvania law’s general exclusion of such evidence in strict liability cases:
Having reached the conclusion that evidence of industry standards relating to the design of the [product], and evidence of its widespread use in the industry, go to the reasonableness of [defendants’] conduct in making its design choice, we further conclude that such evidence would have improperly brought into the case concepts of negligence law. We also conclude that such evidence would have created a strong likelihood of [improperly] diverting the jury’s attention from the [product] to the reasonableness of the [manufacturer’s] conduct in choosing its design.
Lewis v. Coffing Hoist Division, Duff-Norton Co., 515 Pa. 334, 528 A.2d 590, 594 (1987); see also Majdic v. Cincinnati Machine Co., 370 Pa.Super. 611, 537 A.2d 334, 338-39 (1988); Santiago v. Johnson Machine and Press Corp., 834 F.2d 84, 85 (3d Cir.1987) (applying Pennsylvania law).
In sum, while we predict that the Pennsylvania Supreme Court would hold manufacturers liable only for choosing between the safety features feasible at the time the product was manufactured, we predict it would hold them strictly liable for *216that choice. Thus, the manufacturer is liable even if it must choose blindly, with no information about the relative merits of safety features, and even if it exercises all possible care in making that choice. This rule can be justified because it provides manufacturers with an incentive to invest, not only in proven safety features, but also in the testing and development of any new feature that may prove superior. “A strict standard of liability offers the strongest possible incentive, an economic one, for the production of safer products.” J. Beasley, Product Liability and the Unreasonably Dangerous Requirement 61 (1981).
In this case, Clark has conceded that an operator restraint system would have been feasible in 1977. Therefore, in order to determine whether the forklift was “defective”, the only question for the jury was whether an operator restraint system is an “element necessary to make [a forklift] safe for its intended use”, Azzarello, 391 A.2d at 1027, a question that is to be answered on the basis of all the knowledge available at the time of trial. Evidence about what Clark knew or could have known about the desirability of operator restraint systems at the time of manufacture is not relevant to that question. “Since [strict liability is] applicable even to cases in which ‘the seller has exercised all possible care in the preparation and sale of his product,’ the standard of care established by the state of the art is immaterial under strict liability theory.” Beasley, supra, at 393.
Because under Pennsylvania law, evidence of knowledge about the effectiveness of safety devices at the time of manufacture is not probative of whether there is a defect in a crashworthiness case, we would expect it ordinarily to be excluded by the courts of Pennsylvania. Clark contends, in the alternative, however, that the evidence it offered about computer modeling was necessary to rebut evidence the Habeckers offered about the feasibility of operator restraint systems in 1977. As noted above, however, a safety feature’s feasibility is a separate issue from whether it was known to be effective. Clark also notes that the Habeckers provided the first mention of computer modeling, by offering the “Live to Tell About It” safety video. But that video was offered for the proper purpose of showing that Clark, which produced the video, now believes operator restraint systems are effective safety devices. Clark does not explain how it was prejudiced by the video or why it needed to offer rebuttal evidence about computer modeling. Nor does Clark point to any comment by the Habecker’s about Clark’s conduct that would open the door to such evidence. Accordingly, we decline Clark’s invitation to hold that admission of its evidence concerning the lack of computer modeling capability in 1977 was invited error.
C.
We could uphold the judgment for defendants, despite an erroneous evidentiary ruling, if we were convinced that there is a “ ‘high probability’ ” that the error did not prejudice the Habeckers’ substantive rights. McQueeney v. Wilmington Trust Co., 779 F.2d 916, 924 (3d Cir.1985). Thus, the judgment for the defendants should not be set aside if it is highly probable that the jury would have reached the same verdict without hearing the inadmissible evidence. We are not persuaded, however, that the evidence of computer modeling was so benign.
The court properly instructed the jury that it was not to consider Clark’s negligence or lack of negligence in reaching its verdict. Nevertheless, it did not specifically instruct the jury that Clark’s knowledge or lack of knowledge about the desirability of operator restraint systems was irrelevant, and Clark’s closing statement stressed that no one could be certain about the desirability of operator restraint systems until computer modeling was developed, focusing the jury’s attention on the improperly admitted evidence. In arguing that a forklift without such a system was not defective in 1977, Clark stated
you may have heard [plaintiff’s expert] say oh, it was feasible to put a seat belt on the forklift in ’77. It was feasible to *217put a wing on. Sure. Seat belts have been around. They could have thrown a seat belt on the forklift. What would it have cost, nickels and dimes to put a seat belt on? But the point is not trying to cover yourself in litigation or doing what seems to be the simple thing. The question is what is best for the forklift operator. ...
All the [hazards that might come from including an operator restraint system] were considered, and still they couldn’t come up with an answer until the software technology in the computer industry reached a point where they were able to do the kind of computer modeling that you saw on the “Live to Tell About It” video.
App. at 1122a-23a. The court simply instructed the jury that to show defect “the plaintiff must prove by a preponderance of the evidence that an alternative safer design practicable under the circumstances existed.” Without further instruction on what “defective” or “practicable” mean, the jury could easily believe that operator restraints were not “practicable” until the industry knew that they prevented more injuries than they caused. In fact, the district court appears to have relied on that theory in admitting the evidence of computer modeling. Thus, a substantial possibility exists that the jury ruled for Clark in the mistaken belief that the forklift was not defective so long as Clark could not have known that an operator restraint system would make it safer. Accordingly, the judgment for defendants must be reversed.
III.
Because the issue is likely to surface again on retrial, we note that the district court appears to have taken too constricted a view of its own authority on remand. While the parties were preparing for the second trial, the court issued an order stating in relevant part:
the sole issue for retrial is whether the forklift was defective when it left Clark Equipment Company in 1977 because it lacked a seat belt or operator restraint system; ... the parties will be confined to presenting evidence through the witnesses listed in their original pretrial memorandum ...
App. at 162a-63a.
Thereafter, in the course of denying the Habeckers’ request for a continuance to pursue further discovery, the court stated:
... I have been trying to keep this case within the confines of my interpretation of the Third Circuit opinion.
I cannot seem to find anything in [the (sic)] case that would support starting from scratch on this whole thing. I think that I am confined to retry the case on the issue that is left in the case, and that is whether the product was defective for failure to have an operator restraint system.
Nobody has come up with anything to convince me that I can go beyond what the discovery is as of the date of this opinion, that I can allow other issues to come in. Nobody has given me anything to even justify my restricted, and I’ll guarantee you it’s a very restricted view of this opinion as to how I think this second trial should go.
App. at 245a-46a.
The district court was correct in believing that it was bound by our mandate on remand, but it also apparently believed that it had to limit the second trial to the sole issue mentioned in our mandate. We can understand how our mandate could be so read, but it was not our intent to so limit the district court. We referred only to the operator restraint system issue because that was the only issue remaining of those raised by the parties on appeal.
“While a mandate is controlling as to matters within its compass, on remand a lower court is free as to other issues.” Sprague v. Ticonic Bank, 307 U.S. 161, 168, 59 S.Ct. 777, 781, 83 L.Ed. 1184 (1939). For this reason, a district court ordinarily has substantial discretion in conducting further proceedings after a remand.
From the proposition that a trial court must adhere to the decision and mandate of an appellate court there follows the long-settled corollary that upon remand, it may consider, as a matter of first *218impression, those issues not expressly or implicitly disposed of by the appellate decision.... A trial court is therefore free to make any order or direction in further progress of the case, not inconsistent with the decision of the appellate court, as to any question not settled by the decision.
Bankers Trust Co. v. Bethlehem Steel Corp., 761 F.2d 943, 950 (3d Cir.1985); see also Int’l Union UAW v. Mack Trucks, Inc., 917 F.2d 107, 110-11 (3d Cir.1990); Taylor v. U.S., 815 F.2d 249, 252 (3d Cir. 1987). A district court’s discretion therefore includes the authority to allow any theory that would be supported by the complaint and that was not eliminated either by its earlier orders, not pursued on appeal, or by this court’s mandate. For example, the district court’s earlier order dismissing all negligence claims would bar any attempt to revive those claims and this court’s affirmance of the directed verdict for defendant on the throttle issue would bar any attempt to argue liability based on that alleged defect, but other theories can be pursued if the district court concludes that justice so requires.
We express no view on whether the district court should permit a new theory of liability to be pursued at this stage of the proceeding. The district court may well determine that allowing the Habeckers to pursue a theory of liability not raised until after the close of discovery and the completion of an entire trial would result in undue prejudice to the defendants. Moreover, the district court may conclude that Forklifts— which has dismissed its third-party claim against the Army — would be unfairly prejudiced by permission to litigate a new theory. We hold only that the decisions on whether to allow new claims, whether to permit further discovery, and whether to hear additional evidence were all within the district court’s discretion.
IY.
Evidence of computer modeling capabilities in 1977 should not have been admitted into evidence and its admission may have affected the outcome of the proceedings. Accordingly, we will reverse the judgment for the defendants and remand for a new trial.
. They also claim the district court abused its discretion in making specific evidentiary rulings. We have reviewed those rulings and affirm them without further comment.
. In adopting this definition of defect, the Pennsylvania Supreme Court reject the more common approaches that define defect by considering consumer expectations or by balancing the utility and risks of a product. "The decision of this Court in Azzarello sets forth yet [a third] approach to the question of when a design defect may be found to exist.” Lewis v. Coffing Hoist Division, Duff-Norton Co., 515 Pa. 334, 528 A.2d 590, 593 (1987). Whatever the relative merits of these definitions we are bound by that court’s choice.