MURPHY, Circuit Judge.

Rene Junk brought this action in state court on behalf of her son, Tyler (T.J.) Junk, against Terminix International Company (Terminix), Dow Chemical Company and Dow AgroSciences LLC (collectively Dow), and Terminix employee Jim Breneman. Junk alleged that T.J.'s multiple medical conditions were caused by exposure to Dursban, an insecticide manufactured by Dow, distributed by Terminix, and applied to the Junk household by Breneman and other Terminix employees. After Dow removed the case to federal district court, Junk moved to remand for lack of diversity since Junk and defendant Breneman were both citizens of Iowa. The district court denied the motion after concluding that Breneman had been fraudulently joined to evade diversity. Junk's claims against Breneman were subsequently dismissed, and summary judgment was entered for Dow and Terminix. Junk appeals. We affirm the judgment in favor of Dow and Terminix but reverse in respect to the claim against Breneman.

I.

T.J. Junk was born two months premature with an enlarged heart and liver and tachycardia. It was also determined early in his life that T.J. had developmental delay and cerebral palsy. The Junks' home had been infested with spiders during Rene Junk's pregnancy, and she contacted Terminix about the problem. Terminix then sent out its employee Jim Breneman to consult with the family.

Five months before T.J.'s birth, Breneman enrolled the Junks in a Residential Pest Control Service Agreement. Breneman thereafter sprayed the pesticide Dursban inside and outside the Junks' home. Junk alleges that she and her husband had informed Breneman that she was pregnant and had asked him about the safety of Dursban. Dursban contains the synthetic chemical chlorpyrifos. Breneman assured them that Dursban was a "naturally occurring organic material which would have no impact on human beings." The Junks also allege that Breneman and other Terminix employees made similar representations before and after T.J.'s birth. Breneman and other Terminix employees made a total of twenty Dursban applications to the Junk household, the last occurring two years after T.J.'s birth.

On T.J.'s behalf Junk sued Terminix, Dow, and Breneman in Iowa state court.1 Junk alleged that her exposure to Dursban during her pregnancy and T.J.'s exposure to the pesticide after his birth had caused T.J.'s severe neurological problems. Junk claimed breach of express and implied warranties, fraud, negligent misrepresentation, products liability, and negligence. The defendants filed a timely notice of removal to federal district court under 28 U.S.C. §§ 1441, 1446, claiming diversity jurisdiction under 28 U.S.C. § 1332(a).

Junk moved to remand the case to state court under 28 U.S.C. § 1447, arguing that the district court lacked diversity jurisdiction because defendant Breneman and T.J., the party in interest, were both citizens of Iowa. The defendants did not dispute Breneman's Iowa citizenship, but argued that he had been fraudulently joined to defeat diversity jurisdiction. The district court denied Junk's motion to remand the case to state court and her subsequent motions to reconsider. While Junk argued that T.J. had a colorable negligence claim against Breneman due to his false statements and other conduct, the court reasoned that Junk "could not maintain a cause of action ... under Iowa common law" because she had failed to allege explicitly that Breneman had deviated from Terminix's instructions. The court later dismissed Junk's claims against Breneman for failure to state a claim for the same reason.

Discovery continued after Breneman's dismissal, and the district court made three subsequent pretrial evidentiary rulings which Junk contests on appeal. The court first excluded the testimony of Dr. Richard Fenske, one of Junk's expert witnesses. Dr. Fenske is a professor at the University of Washington and an expert on human toxic exposure. The Junks retained him to determine whether T.J. had been exposed to an unsafe level of chlorpyrifos during his mother's pregnancy and after his birth. Dr. Fenske testified that when making toxic exposure and dosage estimates in his research, he usually relies on a "deterministic modeling" method in which he creates an exposure model that accounts for numerous variables. In this case, however, he did not have sufficient data to perform such an analysis. Instead, he compared what he knew about the circumstances of T.J.'s exposure with those in published studies. This comparative analysis led him to conclude that T.J. had been exposed to an unsafe level of chlorpyrifos. Observing that Dr. Fenske had not followed his own usual methodology and concluding that he had relied on a number of ungrounded assumptions in his comparative approach, the district court excluded his opinion on the ground that his methodology was not sufficiently reliable. See Fed.R.Evid. 702; Daubert v. Merrell Dow Pharm., Inc., 509 U.S. 579, 113 S.Ct. 2786, 125 L.Ed.2d 469 (1993).

Dr. Cynthia Bearer's testimony was also excluded. She is a neonatologist and board certified pediatrician whom Junk retained to give her opinion on general and specific causation. She was asked whether chlorpyrifos could have caused the type of injuries that afflict T.J. and whether it did in fact cause them. See Ranes v. Adams Labs., Inc., 778 N.W.2d 677, 688 (Iowa 2010). The district court initially regarded Dr. Bearer's opinions on both general and specific causation to be reliable and admissible, but it indicated that it would revisit the issue after it had examined Dr. Fenske's testimony. That was because Dr. Bearer's opinion on specific causation relied on Dr. Fenske's conclusions. After the court excluded Dr. Fenske's testimony, it found Dr. Bearer's opinion on specific causation to lack a scientific factual basis and declined to admit it.

The district court also granted the defendants' motion in limine to exclude an Environmental Protection Agency (EPA) report that summarized research on the effect of chlorpyrifos exposure on pregnant women, fetuses, and children. After examining a disclaimer on the report that appeared to cast doubt on its trustworthiness, the court expressed concern that the report carried a risk of unfair prejudice and concluded that it was not admissible under Fed.R.Evid. 803(8)(C), 703, or 803(18).

Subsequent to those evidentiary rulings, the district court concluded that Junk could not raise a genuine issue of material fact on the issue of specific causation without Dr. Bearer's opinion. On that basis the court granted summary judgment in favor of Dow and Terminix on all causes of action against them. Junk appeals, challenging the district court's denial of her motion for remand and the subsequent dismissal of the claims against Breneman, the summary judgment for Dow and Terminix, and the orders excluding Junk's experts and the EPA report.

II.

We review the court's denial of remand de novo. In re Prempro Prods. Liab. Litig., 591 F.3d 613, 619 (8th Cir. 2010). A defendant's removal of a case to federal court is appropriate "only if the action originally could have been filed there." Id.; 28 U.S.C. § 1441(a)-(b). A plaintiff may move to remand the case if the district court lacks subject matter jurisdiction. 28 U.S.C. § 1447(c). If the district court concludes that it does not have subject matter jurisdiction, it must remand the case. Id.; see Prempro, 591 F.3d at 620.

Subject matter jurisdiction asserted under 28 U.S.C. § 1332 may be maintained only where there is complete diversity, that is "where no defendant holds citizenship in the same state where any plaintiff holds citizenship." Prempro, 591 F.3d at 620. In an exception to this rule, a district court may retain jurisdiction where the nondiverse defendant has been fraudulently joined. Id. Joinder is fraudulent "when a plaintiff files a frivolous or illegitimate claim against a non-diverse defendant solely to prevent removal." Id.

The parties dispute whether we should review the district court's denial of Junk's remand motion and subsequent dismissal of Breneman under the fraudulent joinder standard outlined in Filla v. Norfolk Southern Railway Co., 336 F.3d 806, 809-11 (8th Cir.2003), or under the standard in Fed.R.Civ.P. 12(b)(6). This question is significant because the two standards differ, with Rule 12(b)(6) being more demanding. Under the Filla standard "the district court's task is limited to determining whether there is arguably a reasonable basis for predicting that the state law might impose liability based upon the facts involved." Filla, 336 F.3d at 811. In contrast, "[t]o survive a [Rule 12(b)(6)] motion to dismiss, a complaint must contain sufficient factual matter, accepted as true, to `state a claim to relief that is plausible on its face.'" Ashcroft v. Iqbal, ___ U.S. ___, 129 S.Ct. 1937, 1949, 173 L.Ed.2d 868 (2009) (quoting Bell Atl. Corp. v. Twombly, 550 U.S. 544, 570, 127 S.Ct. 1955, 167 L.Ed.2d 929 (2007)).

Junk argues that the Filla standard is the proper test to decide whether a remand to state court was appropriate and whether the court's subsequent dismissal of Breneman was proper. This was the approach we followed in Wilkinson v. Shackelford, 478 F.3d 957 (8th Cir.2007). Dow and Terminix argue, however, that a remand order should be reviewed in the same manner as a Rule 12(b)(6) dismissal, citing Simpson v. Thomure, 484 F.3d 1081 (8th Cir.2007).

We do not consider Simpson's holding to be as broad as appellees suggest. In Simpson a nondiverse defendant opposed remand to state court since he was immune from suit under a state workers compensation law. 484 F.3d at 1083. We chose not to apply the Filla fraudulent joinder standard in that situation because the question there turned on the issue of immunity. Id. at 1084 n. 2. Here there is no contention that Breneman is immune from suit in Iowa, and Simpson does not apply. Appellees' overly broad reading of that case would effectively reverse our practice of examining the question of fraudulent joinder under the Filla standard rather than under Rule 12(b)(6).

In Wilkinson, a case decided before Simpson, we faced facts similar to those present here. The district court decided there that a nondiverse defendant had been fraudulently joined by the plaintiff to defeat diversity jurisdiction. The court therefore denied the plaintiff's motion to remand and granted the nondiverse defendant's motion for dismissal. Wilkinson, 478 F.3d at 960. On the plaintiff's appeal, we applied the Filla standard to address the fraudulent joinder question first. Id. at 963. After concluding that the district court had erred in holding there had been fraudulent joinder, we reversed the dismissal of the nondiverse defendant and instructed the court to remand the claim to state court. Id. at 964.

We took a similar approach in Menz v. New Holland North America, Inc., 440 F.3d 1002 (8th Cir.2006). In Menz, the district court had denied a remand motion based on fraudulent joinder and later dismissed the plaintiff's claims. Id. at 1004. We reviewed the court's fraudulent joinder analysis under the Filla standard. Id. Only after holding that the nondiverse defendant had been fraudulently joined, did we then review the court's subsequent dismissal order. Id. at 1005-06.

A comparison of these decisions shows no direct conflict between Simpson and Wilkinson. Simpson involved a nondiverse defendant who had been fraudulently joined since he claimed immunity under a state workers compensation law. Neither Wilkinson nor the case before the court have such an immunity issue. Moreover, even if there had been a conflict between Wilkinson and Simpson, "the better practice normally [would be] to follow the earliest opinion, as it should have controlled the subsequent panels that created the conflict." United States v. Robertson, 606 F.3d 943, 949 (8th Cir.2010).2 For these reasons our Wilkinson approach should be followed here. We therefore review the district court's denial of remand under the Filla standard and would then reverse it along with Breneman's dismissal if we were to conclude that he was not fraudulently joined.

Joinder is not fraudulent where "there is arguably a reasonable basis for predicting that the state law might impose liability based upon the facts involved." Filla, 336 F.3d at 811. That question turns on whether Junk might have had a "colorable" claim against Breneman, "not on the artfulness of [Junk's] pleadings." Wilkinson, 478 F.3d at 964; see Menz, 440 F.3d at 1005. "All doubts about federal jurisdiction should be resolved in favor of remand to state court." Prempro, 591 F.3d at 620. In considering whether Breneman was fraudulently joined, the district court concluded that Junk could not maintain any of the causes of action asserted against him. On appeal, Junk argues only that the court erred in its analysis of the negligence claim against him.

An employee may be held liable under Iowa law for on the job negligence. Hartig v. Francois, 519 N.W.2d 393, 395 (Iowa 1994). There are limits to such liability, however. "An employee who carefully carries out the plans, specifications and directions given the employee by the employer is not liable for defects in those plans, specifications or directions unless they are `so obviously defective and dangerous that no reasonable man would follow them.'" Hartig v. Francois, 562 N.W.2d 427, 431 (Iowa 1997) (quoting W. Page Keeton et al., Prosser and Keeton on the Law of Torts § 104A (5th ed.1984)). The district court cited Hartig in ruling that Junk's complaint failed to allege that Breneman "deviated from the instructions given by his employer, Terminix, regarding sales and warnings for the pest control services."

Although Junk's complaint was not as artful as it might have been, we conclude that it stated facts sufficient to make a colorable claim against Breneman under Iowa law. Junk alleged that Breneman "made representations to [the Junks] that Dursban was a naturally occurring organic material which would have no impact on human beings." In the section of Junk's complaint alleging negligence, Junk alleged that Terminix and Breneman "[failed] to warn ... of health risks associated with the use of [Dursban]" and did "not requir[e] consumers to vacate their house when Dursban was sprayed." Although the conjunctive nature of Junk's complaint did not always distinguish between allegations against Terminix, the employer, and Breneman, the employee, "all doubts ... should be resolved in favor of remand." Prempro, 591 F.3d at 620. Junk's complaint may reasonably be read to raise the issue whether Breneman acted outside of the scope of Terminix's "plans, specifications and directions."3 See Hartig, 562 N.W.2d at 431. For example, it may be reasonably inferred from the complaint that Terminix would likely not have instructed its employees to represent that Dursban was "a naturally occurring organic material" and "would have no impact on human beings" when Dursban was in fact a synthetic chemical.

We conclude that there was "arguably a reasonable basis for predicting that the state law might impose liability based upon the facts involved." Filla, 336 F.3d at 811. This conclusion is "all that is required to defeat a fraudulent joinder challenge." Wilkinson, 478 F.3d at 964. The district court's order denying remand and its subsequent dismissal of Junk's claim against Breneman should therefore be reversed and that claim remanded to state court.

III.

After the district court denied the motion for remand on April 3 and dismissed Breneman on October 3, 2006, the litigation proceeded against the corporate defendants throughout 2007 and 2008. During that time the court issued orders excluding evidence and it granted summary judgment in November 2008 in favor of Dow and Terminix on all remaining claims. Junk urges that the erroneous denial of remand should void the court's subsequent rulings in favor of Dow and Terminix, but "a district court's error in failing to remand a case improperly removed is not fatal to the ensuing adjudication if federal jurisdictional requirements are met at the time judgment is entered." Caterpillar Inc. v. Lewis, 519 U.S. 61, 64, 117 S.Ct. 467, 136 L.Ed.2d 437 (1996). Upon Breneman's dismissal, the court's diversity jurisdiction was perfected and the litigation could proceed as to Dow and Terminix. See Wilkinson, 478 F.3d at 964 n. 4.

Because the district court's summary judgment order turned on disputed evidentiary rulings, we address those rulings first. We reverse a trial court's decision on the admissibility of expert evidence only on the basis of a clear and prejudicial abuse of discretion. Barrett v. Rhodia, Inc., 606 F.3d 975, 980 (8th Cir.2010). Rule 702 provides:

In order to be admissible scientific testimony must be "both reliable and relevant." Barrett, 606 F.3d at 980.

To be deemed reliable, the methodology underlying an expert's conclusions must be "scientifically valid." Id. Speculative testimony should not be admitted. Concord Boat Corp. v. Brunswick Corp., 207 F.3d 1039, 1057 (8th Cir.2000). In deciding to exclude expert testimony, "[a] court may conclude that there is simply too great an analytical gap between the data and the opinion proffered." Gen. Elec. Co. v. Joiner, 522 U.S. 136, 146, 118 S.Ct. 512, 139 L.Ed.2d 508 (1997).

Dr. Richard Fenske was retained by Junk to offer his opinion on T.J.'s exposure to chlorpyrifos during his mother's pregnancy and after his birth. Dr. Fenske's academic experience and expertise are not in dispute. The district court concluded, however, that Dr. Fenske's methodological approach in this particular case was not sufficiently reliable. Dr. Fenske testified that he generally employs a deterministic modeling approach to estimate toxic exposure levels. Since he lacked the data necessary to conduct such an analysis here, he could not estimate exposure levels as he normally does. Instead he resorted to a comparative analysis, analogizing to previous studies of household chlorpyrifos exposure before ultimately concluding that T.J. had been exposed to an unsafe level of the chemical.

The district court identified several grounds for its decision that Dr. Fenske's opinion was not sufficiently reliable: Dr. Fenske admitted that he had been unable to follow the modeling methods that he uses in his published research and teaching because he lacked necessary data. In this case he used a comparative method instead, comparing the circumstances the Junks experienced to several studies which had measured the effects of chlorpyrifos exposure.

The court concluded that Dr. Fenske's comparative analysis depended on unsupported assumptions. He did not account for differences between conditions in the Junk household and those described in the articles he consulted. In one instance his only basis for comparison was the fact that the Junk household and those in a particular study were all treated with chlorpyrifos. In another he relied on a study where the only common variable between the Junks' experience and the homes studied was the total amount of chlorpyrifos applied. Dr. Fenske disregarded other important variables such as where and how chlorpyrifos was applied in the household and whether the homes in a comparison study were the same size as the Junks' home.

While Junk correctly notes that Dr. Fenske was not required to produce "a mathematically precise table equating levels of exposure with levels of harm," Wright v. Willamette Indus., Inc., 91 F.3d 1105, 1108 (8th Cir.1996), the district court did not apply such a high standard. Rather, the court determined that Dr. Fenske had not used a "scientifically valid" method to estimate that T.J.'s exposure exceeded a safe level. We agree with the district court that Dr. Fenske's failure to follow his own general practice and his reliance on unfounded assumptions in his comparative method created "too great an analytical gap" between his opinion and the data on which it relied. Joiner, 522 U.S. at 146, 118 S.Ct. 512. For these reasons we conclude that the district court did not abuse its discretion by excluding Dr. Fenske's expert opinion on chlorpyrifos exposure.

Junk also appeals the district court's exclusion of Dr. Bearer's expert opinion on specific causation. Her opinion was based on her differential diagnosis of the cause of T.J.'s neurological condition. Although the court initially planned to receive it, the court indicated that her "expert causation opinion [would] be reexamined... when the admissibility of the expert testimony of Dr. Fenske [came] before the Court" since she had relied on his conclusions. After the district court addressed Dr. Fenske's opinion on exposure levels and found it wanting, it revisited Dr. Bearer's testimony and excluded her opinion on specific causation.

Because Dr. Bearer's differential diagnosis depended on Dr. Fenske's opinion on exposure, the district court did not abuse its discretion in excluding it. A differential diagnosis begins with an expert's "ruling in" plausible causes of an injury. See Kudabeck v. Kroger Co., 338 F.3d 856, 860-61 (8th Cir.2003). Then the expert "rules out" less likely causes until the most likely cause remains. See id. While differential diagnoses are generally admissible, they should be excluded if they are scientifically invalid. Id. at 861. Dr. Bearer relied on Dr. Fenske's analysis for the proposition that T.J. had been exposed to an unsafe level of chlorpyrifos. After the district court excluded Dr. Fenske's opinion, Dr. Bearer had no valid basis on which to hold chlorpyrifos exposure to be a cause of T.J.'s medical problems. Without a scientific basis for including unsafe chlorpyrifos exposure in her differential diagnosis, Dr. Bearer's opinion amounted to speculation. The district court did not abuse its discretion by excluding Dr. Bearer's opinion on specific causation.4

Junk further appeals the district court's order excluding an EPA report she had offered. The report summarized the findings of various studies. Junk argues that the report was admissible either under Rule 803(8)(C) or as data relied upon by an expert under Rule 703. We conclude that the district court correctly determined that the exceptions provided by those rules do not apply to this report. Rule 803(8)(C) provides an exception to the hearsay rule permitting admission of "factual findings resulting from an investigation made pursuant to authority granted by law ... unless the sources of information or other circumstances indicate a lack of trustworthiness."

While the parties contest whether the EPA report was "made pursuant to authority granted by law," we need not reach that issue. To be admissible under Rule 803(8)(C), "EPA reports must survive a trustworthiness inquiry." O'Dell v. Hercules, Inc., 904 F.2d 1194, 1206 (8th Cir. 1990). A prominent disclaimer on the report states that the "information provided does not necessarily reflect the views of the [EPA], and no official endorsement should be inferred." The disclaimer further states that the report is "not sufficiently detailed nor is it intended to be used directly for environmental assessments or decision making." Whether or not the report "result[ed] from an investigation made pursuant to authority granted by law," the district court reasonably exercised its discretion in determining that this disclaimer indicated "a lack of trustworthiness." See id. at 1206-07 (holding that district court did not abuse its discretion by excluding government reports that "explicitly qualified" their conclusions).

Junk's argument that the EPA report is admissible through Dr. Bearer's expert testimony under Rule 703 also fails. Rule 703 provides that "[f]acts or data that are otherwise inadmissible shall not be disclosed to the jury by the proponent of the opinion ... unless the court determines that their probative value in assisting the jury to evaluate the expert's opinion substantially outweighs their prejudicial effect." Because the report takes the form of a brief summary of research rather than a scholarly publication, the court could determine that it had little probative value. The district court also concluded that the report's unfair prejudicial effect could be substantial if the jury were misled "into believing that [the] report records findings by the EPA." Because we agree that the report's probative value was minimal and its prejudicial potential substantial, we conclude that the district court did not abuse its discretion in excluding it.

IV.

After making its evidentiary rulings, the district court entered an order granting summary judgment to Dow and Terminix on all claims. Junk appeals from that judgment. "We review a grant of summary judgment de novo, affirming if the record shows that there is no genuine issue of material fact and the prevailing party is entitled to judgment as a matter of law." Barrett, 606 F.3d at 983-84. Iowa tort law governs this diversity action. See id. at 984.

To prevail in a toxic tort case such as this, the plaintiff must show both general and specific causation. Ranes, 778 N.W.2d at 688. "General causation is a showing that the drug or chemical is capable of causing the type of harm from which the plaintiff suffers.... Specific causation is evidence that the drug or chemical in fact caused the harm from which the plaintiff suffers." Id. In proving both types of causation, "expert medical and toxicological testimony is unquestionably required to assist the jury." Id. To succeed in her claims, Junk needed to present expert testimony showing that the chlorpyrifos could have caused T.J.'s injuries and that it did in fact cause those injuries.

Junk's only expert witness on specific causation, Dr. Bearer, did not survive the district court's Daubert analysis. After the court properly excluded Dr. Bearer's testimony, Junk could not prove specific causation as required under Iowa law. As there was no longer a genuine issue of material fact as to that necessary element, Dow and Terminix were entitled to judgment as a matter of law and the court properly granted summary judgment.

Finally, Junk argues that the district court violated her Seventh Amendment right to a jury trial by improperly weighing evidence in the course of its Daubert rulings. As already discussed, we conclude that the court did not abuse its discretion in excluding unreliable expert testimony. Junk does not cite any case for the notion that a proper Daubert ruling violates a party's right to a jury trial. Nor does a grant of summary judgment violate a party's right to a jury trial where, as here, the moving party is entitled to judgment as a matter of law. See Harris v. Interstate Brands Corp., 348 F.3d 761, 762 (8th Cir.2003). Junk's constitutional claim lacks legal merit.

V.

For these reasons we affirm the summary judgment in favor of Dow and Terminix, but reverse the order dismissing Junk's claim against Breneman and return that claim to the district court with instructions to remand it to state court.