CRM in the Cockpit: An Analysis of Crew Communication in the Crash of United Airlines Flight 232

Abstract

This study presents an analysis of flight crew communication during the crash of United Airlines Flight 232 at Sioux Gateway Airport in Iowa, USA. Conversation analysis (CA) techniques are used to identify five recurring phenomena in the crew communication and five critical interactions. These are combined to produce a description of the communication process during an unprecedented airline emergency. One of the findings is that communication was simplified and the pilots largely used plain language when speaking with air traffic control (ATC). This was an appropriate communication strategy for the context of the Flight 232 accident but would be problematic if applied to other situations. The analysis also identifies aspects of the crew’s performance that are relevant to contemporary crew resource management (CRM) programs: active participation in communication, updating the shared mental model, making problem solving a joint task, expanding the team boundary to accept an off-duty pilot, and managing the workload. Finally, the study highlights significant details of the Flight 232 accident that are often overlooked and may not generalize to other settings.

1. Introduction

During the 1970s, there were a series of high-profile accidents in which airline flight crews were unable to effectively manage the resources at their disposal. These accidents included: the 1972 crash of Eastern Air Lines Flight 401 in the Florida Everglades; the 1977 runway collision between KLM Flight 4805 and Pan Am Flight 1736 at Tenerife; and the 1978 crash of United Airlines Flight 173 near Portland, Oregon. In the last of these, the United Airlines crew became so absorbed with a landing gear problem that they did not realize their fuel was running out. Each accident was complex and unique, but they all featured inadequate leadership and a poorly functioning team that did not make effective use of the available information and human resources, with tragic consequences.

In response to these and other accidents, NASA organized a three-day workshop for researchers and industry representatives in 1979 to discuss the concept of resource management on the flight deck. In his opening presentation, Chairman John Lauber provided a review of relevant research: interviews conducted by NASA with airline pilots; a simulator study involving flight crews from an American airline; a study of 62 airline accidents; and a study of 250 jet transport incidents [1]. This seminal conference led to the creation of cockpit resource management (CRM) training programs, which were soon introduced by airlines in the United States and subsequently spread worldwide.

Early CRM programs attempted to correct problems such as overly authoritarian behavior by captains and a lack of assertiveness by junior crew members. They were based on existing management training courses, and often used lectures and video presentations, as well as team tasks such as the NASA moon survival game [2].

The most comprehensive of the early programs was that introduced by United Airlines in 1980, which was called Command Leadership Resource Management. Developed as a collaboration between airline training staff, members of the Air Line Pilots Association (ALPA) and two social psychologists, it helped participants to reflect on their personal managerial styles. A key point of the United Airlines program was that courses were held during both initial and recurrent training, allowing for the reinforcement of CRM concepts. In addition to roleplays of flight scenarios, it featured line-oriented flight training (LOFT) in which crews conducted flight segments in a simulator. Video recordings of crew interactions were used in the post-session debriefings [3].

In the mid-1980s, NASA held another CRM workshop for industry. The focus had now switched to team building and there was a change in terminology from Cockpit Resource Management to Crew Resource Management. This change was intended to emphasize team dynamics and interactions with personnel outside the cockpit, such as cabin crew, dispatchers, and air traffic control (ATC) [4].

CRM training subsequently evolved through several more generations, but the fundamental goal has remained unchanged: to train airline crews in techniques that enable them to work as effective teams and avoid problematic behavior patterns. Contemporary CRM programs typically include training in the following skill areas: communication/interpersonal skills, situation awareness, problem-solving/decision-making/judgement, leadership/followership, stress management, and critique [5]. Outside of aviation, CRM training has been adopted by a number of other high-reliability industries such as healthcare, firefighting services, nuclear power generation, maritime and rail transport, and the offshore oil and gas industries [6,7].

This study focuses on an iconic accident that occurred in 1989, a decade after the NASA conference. It involved an experienced flight crew that had participated in the United Airlines training program. The communication of the crew during the accident is examined to identify how CRM techniques helped them deal with an unprecedented in-flight emergency.

2. The Accident

On 19th July 1989, United Airlines Flight 232 was scheduled to fly from Denver, Colorado, to Chicago, Illinois, and then on to Philadelphia, Pennsylvania. The aircraft was a McDonnell Douglas DC-10-10 wide-body jet with the registration N1819U. As was common at that time, the flight deck crew consisted of three people: the captain, first officer (FO), and flight engineer (FE). There were eight flight attendants (FAs) in the cabin crew, and the plane was carrying 285 passengers.

After taking off at 1409 Central Daylight Time (CDT) from Stapleton International Airport in Denver, the jet climbed to its cruising altitude of 37,000 feet. Just over an hour into the flight, at 1516:10, there was a loud bang followed by vibration of the airframe. The tail-mounted number two engine had catastrophically failed. This caused extensive damage to the tail section, and as a result the aircraft lost all of its hydraulic-powered flight controls. At 1520, the pilots declared an emergency, and, a few minutes later, they were given vectors for Sioux City, Iowa. An off-duty United Airlines training/check captain, who was on the flight as a passenger, offered assistance and entered the cockpit at about 1529. Using differential thrust from the remaining two engines, the flight crew compensated for the lack of flight controls and managed to direct the aircraft to Sioux Gateway Airport. Controlling the aircraft was extremely difficult without flight controls, and this was compounded by the phugoid motion it had adopted, which produced a cyclical variation in the pitch, airspeed and altitude, with a period of about one minute. After landing heavily at 1600, the plane skidded off the runway and rolled into an inverted position. The impact forces and subsequent fire destroyed the aircraft, and one flight attendant and 111 passengers were killed. Remarkably, there were 184 survivors. Figure 1 shows a timeline of the accident.

The accident investigation was carried out by the National Transportation Safety Board (NTSB) and a 126-page report was published in November 1990. A large part of the report details the engine failure and the manufacturing process that was used to make fan disks for stage one of the engine. Investigators found that an undetected fatigue crack had developed from a metallurgical defect in a critical area of a fan disk, and this crack led to the in-flight disintegration of the disk [8].

During the investigation, flight simulator studies were conducted to recreate the accident scenario. In the simulator sessions, crews of DC-10 pilots and instructors attempted to control the airplane and land it safely, without the hydraulic flight control systems and using only two of the three engines. It was found that crews could control individual flight parameters (e.g., direction, attitude, speed, vertical speed, or touchdown point) but were unable to control all of the parameters simultaneously. The NTSB concluded that “under the circumstances the UAL flight crew performance was highly commendable and greatly exceeded reasonable expectations” [8] (p. 76).

As shown in

Table 1. The flight crew of United Airlines Flight 232 [8].

	Captain (CAPT)	First Officer (FO)	Flight Engineer (FE)	Check Captain (CAPT 2)
Age	57	48	51	46
Flight hours (total)	29,967	about 20,000	about 15,000	about 23,000
Flight hours (DC-10)	7190	665	33	2987

, the crew of United Flight 232 were very experienced: the captain, FO, and FE had all accumulated between 15,000 and 30,000 flight hours. The training/check captain who joined them during the emergency, and whose work included flight instruction and pilot proficiency checks for DC-10 aircraft, also had a large number of flight hours.

Although CRM training is only mentioned in one sentence of the accident report, in the years that followed, United Flight 232 came to be regarded as a validation of CRM techniques. Captain Al Haynes gave presentations about the accident in which he commended the United Airlines’ CRM program. He also discussed the factors that he thought enabled so many people to survive the crash: luck, communications, preparation, execution, and cooperation. In terms of luck, he noted that the location of the plane, the weather conditions, the time of day, and the particular day of the month were all significant in helping the crew land at Sioux Gateway Airport and in augmenting the support they received from emergency services [9,10].

3. Materials and Methods

3.1. Methodology

This study is in line with contemporary approaches to the analysis of accidents in high-risk complex systems, such as Reason’s Swiss Cheese model, the Human Factors Analysis and Classification System (HFACS), and the Man–Technology–Organization (MTO) model. These approaches acknowledge that accidents are invariably the result of the co-incidence of multiple factors operating at various levels, from the acts of individuals to the influence of organizational culture or regulations [11,12].

The study was conducted by the author using a three-stage process, which has previously been used to analyze flight crew communication during the crash of Japan Airlines Flight 1045 in Anchorage, Alaska [13]. The methodology was developed from the conversation analysis (CA) tool that Nevile created for the Australian Transport Safety Bureau [14]. This tool includes protocols for transcribing and analyzing recorded voice data, and can be used in investigations of aviation or other transport accidents. It enables an analyst to identify both recurring communication phenomena of special interest and key periods of interaction that warrant close analysis.

Audio recordings were not available for this analysis; the NTSB can publish excerpts of CVR transcripts that are relevant to an investigation, but is prohibited from releasing audio recordings under US law (49 U.S. Code § 1114). Instead, the methodology was applied to the cockpit voice recorder (CVR) transcript. The transcript data were transferred to an Excel spreadsheet file, and, to facilitate the analysis, they were divided into separate tracks for intra-cockpit dialogue and air–ground radio communications.

The first stage involved quantifying the speaker source distribution. The number of turns was calculated for each of the discourse participants, with separate counts for intra-cockpit dialogue and radio transmissions.

In the second stage, the transcript data were examined using the CA techniques of unmotivated looking, in which there is no specific focus or intention, and motivated looking, which is informed by knowledge about the context, such as the findings of the accident report [15,16]. This allowed the following features to be identified:

• Recurring phenomena (RP): communication phenomena that recur over the whole recording and are of particular interest;
• Critical interactions (CIs): key periods of interaction that were pivotal to the outcome of the event and therefore require close analysis.

The data were first interrogated using an iterative process of unmotivated looking. Over several cycles, new communication features were added and the existing ones were refined. The iterations continued until the recurring phenomena and critical interactions had stabilized, and no further instances of the recurring phenomena were identified. The data were then examined using the technique of motivated looking. This served as a final check for communication features that might have been previously overlooked.

The third stage involved the analysis of the recurring phenomena and critical interactions, which drew on a range of approaches and areas of study including conversation analysis, interdiscourse communication, applied linguistics, pragmatics, sociolinguistics, and aviation human factors. The rationale for using multiple approaches was that, by examining the same event from different angles, it is possible to clarify both expected and unexpected findings. The layering of information provides a richer explanation of the causal processes at work during the accident.

A limitation of this methodology is that it uses a CVR transcript instead of the original audio. The process of transcribing audio for an accident investigation is much simpler than CA transcription, and, as a result, there is a significant loss of detail. For instance, accident report transcripts usually do not indicate the precise timing of pauses and overlaps, and they do not include information about speech delivery characteristics such as rising or falling intonation.

On the other hand, a key advantage of the methodology is that CA techniques allow interactions to be examined as they unfolded, utterance by utterance. This is a defense against hindsight bias. It enables an analyst to make sense of an accident while bearing in mind that the participants did not have access to all of the information about their situation and did not know what the outcome would be. As the pilot and academic Dekker has noted: “You must guard yourself against mixing your reality with the reality of the people you are investigating.” [17] (p. 27).

3.2. CVR Transcript

The NTSB report states that the CVR recording began at 1526:42 and covered the final 33 minutes 34 seconds of the flight. The accident report does not include the transcript but simply has a three-page summary with selected excerpts. A copy of the transcript was instead downloaded from the Aviation Safety Network website, which is managed by the Flight Safety Foundation [18]. The downloaded transcript data were transferred to an Excel file and divided into three separate tracks:

• Intra-cockpit dialogue between the captain, FO, FE, check captain, and one flight attendant;
• Radio transmissions from the captain, FO, and Sioux City approach controller (APPR);
• Radio transmissions from the FE, the United Airlines dispatch facility (DISP), and the United Airlines maintenance facility (MAINT).

The transcript has several shortcomings. First, it is incomplete: there are six parts where a brief comment indicates that dialogue has been omitted. Second, there are discrepancies between the transcript and excerpts quoted in the NTSB report. Third, a note in the transcript implies there were two jumpseat pilots (the “jumpseat captain” and “jumpseat training pilot”) whereas the NTSB report indicates that these were actually the same person (i.e., the training/check captain). Finally, the transcript does not include any time information; therefore, where possible, times were added by cross-referencing with the CVR summary in the report. The resulting transcript was used for the analysis.

4. Results

4.1. Speaker Source Distribution

The number of turns for each of the discourse participants is shown in

Table 2. Distribution of turns by discourse participants.

	A. Intra-Cockpit Dialogue	B. Radio Transmissions	A + B	A + B (%)
CAPT	129	47	176	36.8%
FO	54	15	69	14.4%
FE	25	22	47	9.8%
CAPT 2	82	0	82	17.2%
FA	5	0	5	1.0%
APPR	0	60	60	12.6%
MAINT	0	23	23	4.8%
DISP	0	3	3	0.6%
Unknown	13	0	13	2.7%
Total	308	170	478	100.0%

. Noting that several sections are omitted in the transcript, the remaining CVR dialogue amounts to 478 turns. The captain produced the most utterances (36.8%), followed by the check captain (17.2%), FO (14.4%), approach controller (12.6%), and FE (9.8%). Intra-cockpit dialogue accounted for 64.4% of the turns, while 35.6% were radio transmissions. Turn length ranged from single-token utterances (e.g., “Left”, “Yep”, “Okay” by the captain) to a 120-token turn by the FE in which he described the engine failure to the maintenance facility.

4.2. Recurring Phenomena (RP)

Five recurring communication phenomena were identified in the CVR transcript. The first two phenomena, role switching and plain language use, are found in radio transmissions between the pilots and the approach controller. The next two phenomena, sharing information about aircraft damage and repeated requests for the same information, occur in both the intra-cockpit dialogue and radio exchanges. The final phenomenon, humor, is—with one notable exception—confined to intra-cockpit dialogue among the flight crew.

4.2.1. RP 1: Role Switching by the Captain and FO

The first communication phenomenon is the role switching (between the captain and FO) of the pilot who speaks with the approach controller. At the start of the CVR transcript, the captain is responsible for radio contact with the controller. However, in the following 30 min, there are 14 instances of the FO assuming this role. In two cases, the FO contacts the controller to request information about the heading and distance to Sioux Gateway Airport. In 10 instances, the controller initiates a short exchange (two or three turns) and the FO replies in place of the captain.

Table 3. Exchange between the approach controller (APPR) and FO at about 1547.

Turn	Speaker	Content
1	APPR	United two thirty-two heavy, there is a small airport at twelve o’clock and seven miles. The runway is four thousand feet long there.
2	FO	Hey, I’m controllin’ it myself now. As soon as the captain gets back on, he’ll give me a hand here. He’s talking on the PA.

shows an example, with a transmission from the approach controller giving information about a small airport. The FO responds, explaining that the captain is currently making an announcement to passengers.

4.2.2. RP 2: Plain Language Use in Pilot–ATC Communication

The second recurring phenomenon is the extensive use of plain language by the captain and FO in radio transmissions to the controller. Pilot–ATC communication consists of two varieties of language: standard phraseology and plain language. Standard phraseology (also called radiotelephony or RT phraseology) is designed for routine flight operations and has a number of distinct features: pronunciation protocols, a limited lexicon, modified syntax, standardized message structures and specific exchange patterns [19,20].

Plain language is meant to be used for non-routine operations, and may be defined as any language used by pilots or controllers that is not standard phraseology. The essential benefit of plain language is that it allows innovative language use to deal with unexpected situations not covered by phraseology. Although sometimes considered to be synonymous with general language or conversational language, it is constrained by regulations. Even when using plain language, pilot–ATC communications are required to be clear, concise, and unambiguous [19,21].

The pilots used plain language in almost all of their radio transmissions with the controller. Table 7, which shows the first critical interaction, illustrates a number of features of plain language that distinguish it from phraseology:

• Additive discourse: in turn 1, the captain establishes a topic (lack of controllability) and then adds a series of short statements giving extra information;
• Untidiness: in turn 1, the captain makes contradictory statements about left and right turns, but the controller is able to retrieve the intended meaning and in turn 2 correctly states “you can only make right turns“;
• Fillers: both the captain and controller use the filler “ah” to provide time to compose the next section of discourse, with some fillers at possible turn completion points signaling the speaker wants to continue talking;
• Subject pronouns: “we”, “I”, “you” and “it”;
• Auxiliary verbs: “are”, “do”, “is” and “will”;
• Colloquialisms: “we’re”, “don’t”, “can’t”, “that’s”, “it’s” and “gonna”;
• Vague language: “a little bit”, “maybe” and “very slight”.

The features listed above occur throughout the transcript, but there were some differences between the language use of the captain and the FO. The captain repeatedly used vague language as he tried to explain the difficulty in controlling the aircraft: “almost none”, “pretty much under control”, “a little better control” and “a little bit” (four times). He also used other prefabricated lexical phrases (or chunks) including: “under the circumstances”, “regardless of the condition”, “whatever you do” and “that’s all”. The FO used several idiomatic expressions (two of which are shown in Table 3): “give me a hand”, “gets back on”, “pick it out” and “givin’ it heck”.

4.2.3. RP 3: Sharing Information About Aircraft Damage

In the third pattern, the flight crew share information about damage to the aircraft with each other, the approach controller, dispatch, maintenance, and the cabin crew. There are 10 instances of this phenomenon, all of which occur in the first two-thirds of the transcript.

Table 4. Intra-cockpit dialogue between the captain, FO and FE at 1530.

Turn	Speaker	Content
1	FO	What’s the hydraulic quantity?
2	FE	Down to zero.
3	FO	On all of them?
4	FE	All of them.
5	CAPT	Quantity, quantity is gone?
6	FE	Yeah, all the quantity is gone. All pressure is [gone].

shows one example where the FO initiates a brief discussion with the FE and captain, with the result that all three crew members have correct information about the hydraulics situation.

There are three more instances in the few minutes after the arrival of the check captain in the cockpit, when the captain updates him about the hydraulics, elevator control and the number two engine. In another instance, the captain briefs the senior flight attendant about the emergency landing and explains that they have “almost no control of the airplane”.

The remaining five instances involve the FE contacting the maintenance facility about the loss of engine two, the hydraulic systems, control surfaces, and damage to the tail. These exchanges are often lengthy and are spread across the first two-thirds of the transcript. Finally, after checking that maintenance is unable to provide assistance, the captain instructs the FE to stop transmitting, saying: “Okay, well forget them. Tell ’em you’re leaving the air, and you’re gonna come back up here and help us… and screw ’em.”

4.2.4. RP 4: Repeated Requests for the Same Information

The fourth recurring phenomenon, which occurs 10 times, is for repeated requests to be made for the same information. First, the approach controller on three occasions asks how many people are on board the flight.

Table 5. Exchange between captain and approach controller at about 1536.

Turn	Speaker	Content
1	APPR	United two thirty-two, did you get the souls on board count?
2	CAPT	[Let me] tell you, right now we don’t even have time to call the gal…
3	APPR	Roger.

shows the third exchange, with the captain explaining they are too busy to check this information.

Second, there are four repeated requests from the maintenance and dispatch facilities. Early in the transcript, they twice contact the flight crew to ask again about their landing intentions. With the crew struggling to control the aircraft, the FE responds “ah, we don’t know what we’ll be able to do”. Later, after the crew have decided to head to Sioux City, maintenance calls twice to confirm again the loss of hydraulic fluid and flight controls. The FE replies, “Affirmative, affirmative, affirmative”.

Third, in the final three minutes of the flight there are three occasions when the flight crew ask the approach controller to repeat the wind speed or airport elevation. On the third occasion, the captain indicates that overlapping speech in the cockpit is affecting communication when he tells the controller, “Okay, we’re all three talking at once. Say it again one more time”.

4.2.5. RP 5: Humor

Unlikely as it may seem in the context of an emergency situation, the final recurring pattern is the captain making a statement or initiating a short exchange that triggers laughter in the cockpit. The following are five instances from the intra-cockpit dialogue:

• The captain jokes that he was not assessed on the current emergency during his last simulator check;
• The captain asks the FE if he managed to contact the maintenance facility and the FE replies “Yep. Didn’t get any help.”;
• The captain verbalizes the effort to regain elevator control, repeating “come (on) back” six times, and then says they can deal with the steering later;
• The captain says he wants to turn left to a heading of 180 degrees but the plane is only able to make turns to the right;
• The captain verbalizes the attempts to control the aircraft, chanting “back” and “forward” repeatedly, then saying “Won’t this be a fun landing?”

In each instance the laughter probably signals a partial release of built-up stress in the cockpit. The joking may represent a means of dealing with anxiety arising from a situation that none of the crew have ever experienced before, which is more difficult than anything in simulator training and which the maintenance facility is unable to help with. There is also a sense of the absurdity of the predicament: the crew are in a time-critical situation attempting to land a wide-body jet with 296 people on board, and they can only make turns to the right.

Table 6. Exchange between captain and approach controller at about 1558.

Turn	Speaker	Content
1	APPR	United two thirty-two heavy, the wind’s currently three six zero at one one three sixty at eleven. You’re cleared to land on any runway…
2	CAPT	[Laughter] Roger. [Laughter] You want to be particular and make it a runway, huh?

shows the single instance of humor found in the radio transmissions, which occurs shortly before the landing at Sioux City. The controller and emergency services expect the plane to attempt a landing on runway 31. However, this is not possible and the pilots line up on runway 22, which is closed. When the controller belatedly realizes the change, he gives clearance to land on any runway in turn 1. The incongruity between the precision of the controller’s instruction and the difficulty of controlling the plane causes the captain to laugh and joke about whether they will be able to land on a runway at all. This is a funny riposte at a time of intense psychological stress.

4.3. Critical Interactions (CI)

Five key periods of interaction that were pivotal to the outcome of this event were identified in the transcript. The first critical interaction takes place in the pilot–ATC radio communication, while the rest occur in intra-cockpit dialogue.

4.3.1. CI 1: Captain and Approach Controller

The first critical interaction was a seven-turn radio exchange between the captain and approach controller, which is shown in

Table 7. Exchange between captain and approach controller at 1526.

Turn	Speaker	Content
1	CAPT	Ah, we’re controlling the turns by power. I don’t think we can turn right. I think we can only make left turns. We’re starting a little bit of a left turn right now. Maybe we can only turn right. We can’t turn left.
2	APPR	United two thirty-two heavy, ah, understand you can only make right turns.
3	CAPT	That’s affirmative.
4	APPR	United two thirty-two heavy, roger. Your present track puts you about eight miles north of the airport, sir. And, ah, the only way we can get you around [Runway 31] is a slight left turn with differential power or if you go and jocket it over.
5	CAPT	Roger. Okay, we’re in a right turn now. It’s about the only way we can go. We’ll be able to make very slight turns on final, but right now just… we’re gonna make right turns to whatever heading you want.
6	APPR	United two thirty-two heavy, roger. Ah, right turn, heading two five five.
7	CAPT	Two five.

. This exchange occurs at the start of the CVR recording. The transcript does not show any transmissions that took place prior to this.

As previously noted, the captain makes contradictory statements about left and right turns in turn 1: “I don’t think we can turn right. I think we can only make left turns.” and “Maybe we can only turn right. We can’t turn left.”. In turn 2 the controller gives the aircraft call sign (“United two thirty-two heavy”) followed by a hesitation filler “ah” as he retrieves the intended meaning and states “understand you can only make right turns”. In turn 3 the captain confirms the controller’s understanding is correct. This exchange enables the controller and captain to promptly correct a false hypothesis (“we can only make left turns”). The smooth negotiation of the misunderstanding probably gives the captain confidence that the flight is in safe hands.

The exchange continues in turn 4 with the controller explaining that a small left turn will put the plane on track for Sioux Gateway Airport. The captain replies in turn 5 that they will have to make a right turn to reach the heading instructed by the controller. In turn 6, the controller issues a heading change to 255 degrees, which is partially read back by the captain in turn 7.

4.3.2. CI 2: Captain and Check Captain

The second critical interaction was a five-turn exchange between the captain and the check captain, shown in

Table 8. Intra-cockpit dialogue between captain (CAPT) and check captain (CAPT 2) at about 1538.

Turn	Speaker	Content
1	CAPT	My name’s Al Haynes.
2	CAPT 2	Hi, Al. Denny Fitch.
3	CAPT	How do you do, Denny?
4	CAPT 2	I’ll tell you what. We’ll have a beer when this is all done.
5	CAPT	Well, I don’t drink, but I’ll sure as hell have one. Little right turns, little right turns.

. They introduce themselves in this exchange, even though the check captain had first entered the cockpit almost 10 min previously. The captain uses humor to break the ice and welcome the new member into the flight crew team.

In turn 1 the captain gives a greeting by simply stating his name. The check captain responds with a casual greeting (“Hi, Al”) and states his own name. The captain replies with a more formal greeting in turn 3, acknowledging the check captain’s name (“How do you do, Denny?”). In turn 4, the check captain signals a change in topic (“I’ll tell you what”) and then states that they will drink beer later. The tone is optimistic, with the implicit assumption that they are going to survive this emergency. The check captain uses the plural first person pronoun (“we”) to invoke a shared identity as a member of the flight crew, implying they are now working together and later they will relax together. In turn 5 the captain responds with a negative statement (“Well, I don’t drink…”), setting up an expectation that he is going to reject the drink invitation, underlined by his use of the singular first person pronoun (“I”). However, he then upsets this expectation and emphatically accepts the invitation by saying, “…but I’ll sure as hell have one.” The captain’s word choice (“hell”) provides a semantic link to the emergency they are now experiencing, and his response also carries the sense that the situation is bad enough to drive a teetotaler to drink.

There is no record of laughter in the transcript, but this is another example of (very funny) humor delivered in the most difficult of circumstances. The captain then immediately refocuses attention on the task of controlling the plane (“Little right turns, little right turns”), signaling an end to the exchange and reasserting his authority as pilot in command.

4.3.3. CI 3: Captain, FO, and FA

The third critical communication was a 14-turn dialogue between the captain, FO and senior flight attendant, which is shown in

Table 9. Intra-cockpit dialogue between captain, senior flight attendant (FA), and FO at 1540.

Turn	Speaker	Content
1	CAPT	Everybody ready?
2	CAPT	We have almost no control of the airplane.
3	FO	We have no hydraulics at all.
4	CAPT	It’s gonna be tough, gonna be rough…
5	FA	So we’re gonna evacuate?
6	CAPT	Yeah. Well, we’re gonna have the gear down.
7	FA	Yeah.
8	CAPT	And if we can keep the airplane on the ground and stop standing up, give us a second or two before you evacuate.
9	CAPT	‘Brace’ will be the signal; it’ll be over the PA system –‘Brace, brace, brace.’
10	FA	And that will be to evacuate?
11	CAPT	No, that’ll be to brace for landing.
12	FA	Un huh.
13	CAPT	And then if you have to evacuate you’ll get the command signal to evacuate, but I really have my doubts you’ll see us standing up, honey. Good luck, sweetheart.
14	FA	Thank you too.

. This was a briefing to the senior FA about the emergency landing and evacuation.

In turn 1 the captain starts the briefing by asking if the cabin crew are ready. There is no spoken response to this question, so possibly the FA responded with a nod or other gesture. The captain then uses non-technical expressions to explain the severity of the situation in turns 2 (“almost no control”) and 4 (“tough” and “rough”). The FO interjects a statement about the loss of the hydraulic system in turn 3. In turn 5, the FA asks about evacuation. The captain responds in turns 6 and 8 by explaining that they will try to land with the landing gear intact (“standing up”). In turn 9 he states the signal they will give prior to the emergency landing (“Brace”). The FA misunderstands this in turn 10 as the evacuation signal. The captain immediately corrects the misunderstanding in turn 11. He explains in turn 13 that they will give another signal for evacuation but then expresses doubt as to whether the landing gear will survive the landing (“I really have my doubts you’ll see us standing up”). The negative prognosis is softened by the captain’s use of affectionate address (“honey” and “sweetheart”), and he signals that the briefing has ended by wishing her good luck. In turn 14, the FA acknowledges the end of the briefing with a phatic expression (“Thank you too”).

4.3.4. CI 4: Captain, Check Captain, and FO

The fourth critical interaction was a brief exchange about landing gear involving the captain, check captain and FO, shown in

Table 10. Intra-cockpit dialogue between the captain, check captain, and FO at about 1547.

Turn	Speaker	Content
1	CAPT	Anybody have any ideas about [what to do about the landing gear]? He’s [the Engineer] is talking to Sam.
2	CAPT 2	Yeah, he’s talking to Sam. I gonna alternate-gear you. Maybe that will even help you. If there is no fluid, I don’t know how the outboard ailerons are going to help you.
3	CAPT	How do you we get gear down?
4	CAPT 2	Well, they can freefall. The only thing is, we alternate the gear. We got the [landing gear] doors down?
5	CAPT	Yep.
6	FO	We’re gonna have trouble stopping, too.

. In turn 1, the captain initiates troubleshooting by asking the crew members for suggestions about the landing gear (“Anybody have any ideas about…?”). He also states that the FE is talking on the radio with the San Francisco maintenance facility (“He is talking to Sam”). The check captain confirms in turn 2 that the FE is talking with maintenance. He then says he will use the alternate gear extension lever, before changing the topic to control surfaces (“the outboard ailerons”). In turn 3, the captain rephrases his question about landing gear. The check captain responds in turn 4 that the landing gear will drop by freefall if they use the alternate gear extension lever. He then asks whether the landing gear doors are open. The captain briefly confirms they are open in turn 5 (“Yep”). Then, the FO states in turn 6 that it will be difficult to stop the aircraft after landing. This statement marks a topic change and is the start of a six-turn exchange about the effectiveness of the braking system. As a result, the captain’s attempt to troubleshoot the landing gear issue is unsuccessful and the exchange ends without a decision.

4.3.5. CI 5: Captain, Check Captain, FO and FE

The final critical interaction is another exchange about the landing gear, which ends with a decision. The captain has just instructed the FE to stop radio communication with maintenance, and so this discussion involves the captain, check captain, FO, and FE. The first five turns of the exchange are shown in

Table 11. Intra-cockpit dialogue between captain, check captain, FO, and FE at about 1548.

Turn	Speaker	Content
1	CAPT	Anybody got any idea about pottin’ the gear down right now?
2	CAPT 2	All right, I would. I would suggest…
3	CAPT	Should we free fall it?
4	FO	Yeah, yeah. I got to get out of the way to get the door.
5	CAPT	Put it down.

.

Once again the captain initiates troubleshooting in turn 1 by asking the crew for suggestions about the landing gear (“Anybody got any idea about…?”). The check captain starts to respond with a suggestion but is seemingly interrupted by the captain who asks a closed question in turn 3 (“Should we free fall it?”). The FO gives a preferred response (“Yeah, yeah”) and in turn 5 the captain gives a brief instruction to lower the landing gear (“Put it down”). The transcript omits the next section, but, after further discussion, the FE confirms that the landing gear has been extended at about 1549.

5. Discussion

5.1. The Communication Process

By combining the recurring phenomena (RP) and critical interactions (CI), it is possible to trace the communication process that unfolded during this emergency. The period covered by the CVR transcript (i.e., the last 33 min 34 s of the flight) can be divided into two phases.

The first phase corresponds to the first two-thirds of the transcript (1526–1548), when the flight crew were struggling to control the aircraft, sharing information about the damage, and trying to make sense of what had happened. The workload was high throughout this phase, and the communication was a complex mixture of the following:

• Intra-cockpit dialogue between the captain, FO, FE, check captain, and the lead flight attendant;
• Radio transmissions between the captain, FO, and controller;
• Lengthy radio exchanges between the FE and dispatch/maintenance facilities.

The transcript does not show overlapping speech, but it is highly probable that, in this noisy environment, there was a lot of overlapping talk, increasing the risk of distraction and important information being ignored.

One feature of the pilot–ATC radio communication was fluid role switching between the captain and FO (RP 1). Such blurring of roles entails the risk of other duties being overlooked and key information not being shared, but, in the context of this accident, it was effective workload management that allowed the crew to continue ATC communication while minimizing disruption to other work.

Another feature was that the captain and FO used minimal phraseology in their radio transmissions. Instead, they used plain language, including prefabricated lexical phrases and idiomatic expressions (RP 2). The use of prefabricated forms allows native speakers to produce and decode language in real time without undue cognitive load [22,23]. Not having to switch between two varieties of language (phraseology and plain language) further reduced the cognitive load. In effect, this was a form of task shedding, as the pilots abandoned some communication tasks (e.g., using standard phraseology, giving full readbacks) in order to maintain performance while under high levels of psychological stress and workload [24]. The use of plain language was, in part, necessitated by the need to explain an emergency situation not covered by phraseology. For instance, the captain used vague language to describe the difficulty the pilots were having in controlling the aircraft (CI 1).

However, using plain language increases the risk of miscommunication [25]. For example, a false hypothesis about left turns emerged at the start of the transcript, but it was swiftly corrected in dialogue between the captain and controller (CI 1). The fact that the same controller handled the flight throughout the period of the transcript meant there were no handovers from one sector controller to another, which helped reduce the risk of other false hypotheses developing.

During the first phase, the captain mitigated the risk of misunderstandings by actively sharing information about aircraft damage with other crew members and operators on the ground supporting the flight (RP 3 and CI 3). There were also multiple requests for information to be repeated from the maintenance and dispatch facilities (RP 4), which probably reflected the difficulty they were having in making sense of the situation.

The second phase covers the final third of the transcript (1548–1600), when the aircraft was heading towards Sioux City. Workload remained high, but communication was simplified because the captain had instructed the FE to cease contact with the maintenance facility. The captain and FO continued switching roles in radio communication with the controller (RP 1) and they continued using plain language in their transmissions (RP 2). The four members of the flight crew focused their attention on landing the aircraft at Sioux Gateway Airport, and there were two discussions about how to lower the landing gear (CI 4 and CI 5). The cockpit was still noisy, especially in the final minutes of the flight when the crew made three requests for the controller to repeat wind speed or elevation information (RP 4).

5.2. The Flight Crew and ATC

Luck plays a role in many accidents. In the case of United Flight 232, the in-flight failure of engine number two was a staggering piece of bad luck. On the other hand, as Captain Haynes observed after the accident, the crew were fortunate regarding the accident location, time, day and weather conditions. Two further pieces of good fortune may be added to this list.

First, the accident struck a United Airlines aircraft on a domestic flight, which meant that all communications (both inside the cockpit and on the radio) were in English, the first language of the participants. Captain Haynes later acknowledged the calmness of the Sioux City controller as he provided timely and appropriate information. The controller’s ability to understand the sometimes untidy and idiomatic plain language of the pilots was an important factor in guiding the aircraft to the airport. The aircraft type, the DC-10-10, had a range of 3500 nautical miles and was operated by airlines around the world. Another layer of complexity would have been added if the accident had taken place in Europe, the Caribbean or Asia, with a flight crew and controllers speaking different first languages.

Second, as shown in Table 1, the flight crew of United Flight 232 were very experienced. This alone does not guarantee that a crew will form an effective team. The combination of a highly experienced captain and a highly experienced first officer has been identified as one pattern of non-optimal crew composition, especially in terms of communication [26]. Furthermore, even when the crew members are highly skilled, it takes time for them to develop into an effective team [27]. In the case of the captain, FO and FE, the NTSB report states that they were on the third day of a four-day trip, and they had flown together six times during the previous 90 days. The fact that they had already developed into a functioning team was another factor contributing to the outcome of the event. In addition, the training/check captain who happened to be a passenger on the flight was also a very experienced pilot, although he had been a captain for less than three months. One of the challenges facing Captain Haynes was how to integrate this new member into the existing well-established team in the middle of an unfolding emergency.

5.3. CRM in the Cockpit

The analysis highlights several aspects of the flight crew’s performance that reflect interrelated domains of CRM training. In the first domain, communication, the CVR transcript records the active participation of all four flight crew members. Although the transcript is incomplete, it still reveals a dense communication environment with 478 turns in less than 34 min. The captain and approach controller together accounted for almost half of the utterances. In the landing gear discussions, the captain elicited ideas from other crew members and demonstrated that he was open to suggestions (CI 4 and CI 5). The other members of the flight crew—the FO, FE and check captain—contributed more than 40% of the utterances. The FE was responsible for the fewest number of turns (9.8%). This may be explained by his role in communicating with the dispatch and maintenance facilities, which involved less frequent but longer messages describing the aircraft damage. Furthermore, although the FE had accumulated a large number of flight hours, he only had very limited experience in DC-10 aircraft.

For the second domain, situation awareness, the captain took the lead in sharing information about aircraft damage with other crew members and in keeping the approach controller in the loop (RP 3). He also briefed the senior flight attendant prior to the emergency landing and evacuation (CI 3), and instructed the check captain and FE to make visual inspections of the flight controls and tail section damage. Meanwhile, the FE was in regular radio contact with the dispatch and maintenance facilities. The latter could not provide any additional advice about how to deal with the emergency, but Captain Haynes later acknowledged that they were instrumental in promptly activating the crisis response in Sioux City. Throughout the period covered by the transcript, the captain helped the crew to update their shared mental model of the unfolding emergency by following the CRM maxim to “make all the information needed accessible to all people solving the current problem” [28] (p. 96). Given the noisy cockpit environment and high level of uncertainty about what was happening, it was perhaps inevitable that some participants would make repeated requests for the same information (RP 4). This is part of the process of updating the mental model, and is a defense against false hypotheses, which, as Weick notes, are likely to develop and persist in any crisis situation [29].

The third CRM skill area is problem solving. One key feature of the accident was the ability of the flight crew to troubleshoot an extraordinary emergency situation. The crew suggested various hypotheses, collected data in so far as they could to test the hypotheses, and rejected false hypotheses. Several factors enabled them to do this: a cockpit environment in which crew members were able to freely contribute to the discourse; the captain actively engaging crew members by asking for input and suggestions; and the captain taking suggestions seriously. These factors are illustrated in the discussions about the landing gear issue, which were initiated by the captain asking if anyone had any suggestions (CI 4 and CI 5). In so doing, the captain was following another CRM maxim: to “make problem resolution a joint task” [28] (p. 96).

The final domain is leadership and followership, which includes team building, establishing credibility, and workload management. As noted above, the captain, FO and FE had developed into a functioning team prior to the accident flight. The transcript shows that humor was one of the tools used by the captain to facilitate effective teamwork, relieve tension in the cockpit, and help bond a small group operating under intense pressure (RP 5). The captain broke the ice with a joke when he introduced himself to the check captain, acknowledging the expansion of the team boundary and showing the FO and FE that he accepted the new crew member (CI 2). Crucially, the captain was able to use humor and still maintain cockpit discipline, without distracting the crew from flight-related tasks. He displayed traits typical of a highly effective leader by establishing competence through his actions, disavowing perfection and engaging the crew [30].

The other crew members helped manage the workload in several ways: the check captain asked for instructions and took over throttle control of the two remaining engines; the check captain and FE made visual inspections of the wing and tail; the FO repeatedly took over radio communications with the approach controller when the captain was busy with other tasks; and the FE was responsible for communication with the dispatch and maintenance facilities. The captain and FO also managed their individual workloads by simplifying their language use in radio transmissions, largely dispensing with phraseology and instead using prefabricated forms, idiomatic phrases and vague language. As a result, they did not adhere to the requirement for concise and unambiguous pilot–ATC communication. Nevertheless, their language use was entirely appropriate in the context of this accident: it was an unprecedented emergency that necessitated the use of vague language; their aircraft was the only one on the frequency after the emergency was declared; and the controller was a calm, native speaker of English who was able to fill in the gaps in their messages.

5.4. Other Airline Accidents

While acknowledging that airline accidents are complex events with unique contexts, it is instructive to consider some salient similarities and differences between United Airlines Flight 232 and two other accidents that occurred in the same era: the crashes of Delta Air Lines Flight 1141 and American Airlines Flight 965.

On 31st August 1988, Delta Air Lines Flight 1141 crashed during takeoff with the loss of 14 lives. Like Flight 232, it was a regularly scheduled passenger service on a US domestic route (from Dallas Fort Worth International Airport, Texas, to Salt Lake City International Airport, Utah). Both were summer daytime flights, and neither accident involved any weather-related factors. Flight 1141 also featured a three-person flight crew flying a three-engine jet (albeit a narrow-body Boeing 727). The captain was highly experienced, with approximately 17,000 flight hours. However, the NTSB investigation concluded that he and the first officer failed to maintain adequate cockpit discipline. There was extensive joking and non-pertinent conversation while the aircraft was taxiing, which resulted in the crew attempting to take off without properly configuring the flaps or slats [31]. A key contrast with United Airlines Flight 232 is that the Delta Air Lines crew did not maintain their focus on flight tasks and instead allowed themselves to be distracted by humor and casual conversation during a critical phase of flight. United States law was revised to restrict the release of CVR data following the Delta Flight 1141 accident.

The captain of American Airlines Flight 965 on 20th December 1995 was also highly experienced, having accumulated approximately 13,000 flight hours. Otherwise, though, there were many differences between this accident and Flight 232: American Airlines Flight 965 was an international service from Miami, in the United States, to Cali, in Colombia; it had a two-person flight crew; and they were flying a modern twin-jet Boeing 757 aircraft equipped with a glass cockpit. As the plane was descending towards Cali Airport, it crashed into mountainous terrain, killing 151 passengers and all eight members of the crew. The investigation by Colombia’s Aeronautica Civil determined that the probable cause of the crash was the failure of the flight crew to adequately plan and carry out the approach, compounded by problems with the flight management system. The accident report noted that the routing requested by the American crew did not make sense to the Colombian controller, who said he would have questioned the pilots about this if they had spoken Spanish [32]. A human factors analysis of the accident highlighted ambiguous clearances from the approach controller and failures to clarify ambiguities by the pilots [33]. In contrast to Flight 232, one of the key lessons from the Flight 965 accident is the need for clear and unambiguous pilot-ATC communication when native English speakers are interacting with non-native speakers.

6. Conclusions

In addition to being an extraordinary example of airmanship, United Airlines Flight 232 is an important case study of the application of CRM techniques to an emergency situation. Despite being constrained by the available data, this analysis contributes to our understanding of the accident by providing insights into the communication process in the final stage of the flight. It has identified five recurring phenomena in the communication of the flight crew and five critical interactions that were pivotal to the outcome of the event. These have been combined to produce a description of the communication process, which is outlined in Section 5.1. Building on that description, Section 5.3 discusses aspects of the crew’s performance that are relevant to contemporary CRM programs: communication, situation awareness, problem solving, team building and workload management.

Although the story of United Airlines Flight 232 is well known, Section 5.2 calls attention to details that are often overlooked and which were significant factors in the crew’s response to the emergency. First, the four members of the flight crew were very experienced. Second, all participants in the communication were native speakers of English. Third, the original flight crew had developed into a functioning team prior to the accident flight. Given that United Airlines Flight 232 is still cited as an exemplar of how to apply CRM techniques in an emergency situation, this analysis has the following implications for training programs that use the accident as a case study:

• Key features of the accident context should be highlighted, including the crew’s flight experience and the fact that all participants were native English speakers;
• The critical interactions may be used as a basis for guided discussions about flight crew communication and decision making in recurrent training courses.

Another important finding from the analysis is that, while the communication strategies adopted by the captain and FO were appropriate to the Flight 232 context, they are unlikely to be suitable for other situations. Specifically, the pilots’ use of plain language depended heavily on the approach controller’s ability to understand transmissions that were often untidy, vague, or laden with idioms. In situations involving non-native speakers of English, this could result in miscommunication and the propagation of false hypotheses. In such situations, the use of phraseology combined with simple non-idiomatic plain language is appropriate, as stipulated in the International Civil Aviation Organization (ICAO) requirements [34]. The need for appropriate communication is underscored by the example of American Airlines Flight 965, which is discussed in Section 5.4. There are several implications for CRM practice:

• Pilots should be aware of communication threats and should know mitigation strategies for dealing with the threats;
• Idiomatic and colloquial expressions are a threat to clear communication because they can be difficult for non-native speakers to understand;
• Plain language often includes ambiguous expressions, which are a threat as they may cause confusion.

Finally, the United Flight 232 accident draws attention to the importance of CVR data and the value of CVRs with sufficient duration to record the entirety of a safety event [35,36]. The analysis presented herein provides new insights into the accident, but it was constrained by the available data: the CVR only recorded the final 34 min of the flight; it was not possible to access the CVR audio; the accident report does not include the transcript; and the transcript obtained for the analysis lacked time information and omitted several sections of dialogue. More detailed CVR data (including, for example, prosodic information and timing cues) would have permitted a more thorough analysis. Accident investigations vary in the level of detail with which CVR audio is transcribed, as well as in the amount of transcript data included in reports. If transcripts with a greater level of detail are made available, the analyses will yield deeper insights, which can, in turn, be fed back into airline training programs to improve the safety of civil aviation.