The last time Hackerfall tried to access this page, it returned a not found error. A cached version of the page is below, or click here to continue anyway

Breaking down a Ford recall with stats

The history

You know what they say about lemons ... when life gives you lemons, break out your reliability statistics textbook and analyze that lemon!

My lemon is a 2010 Ford Fusion. It has a pretty fast V6 and I actually really like my lemon-Fusion despite its bitterness.

The first week after bringing it home the entire electonics system failed and it had to be taken away on a flatbed --- but hey \()/, electronics are good-as-new type repairs and this is just the beginning of the bathtub curve, right?

Then there was the time three lug nuts sheared in half. The guy at the tire shop hadn't ever seen that happen. Thankfully, Ford announced a recall for that issue.

I think I drove around like this for at least two days before noticing.

Then the car died on the interstate a few times. That turned out to be a faulty throttle body. Again, another recall.

As a consumer, I'm thankful for recalls. As a statistician with a focus on reliability statistics, I can appreciate the role and challenge of engineering and designing vehicles for safety and reliabiity. I grew up south of Flint, Michigan. My dad's an engineer at one of the Big Three. Growing up, both my neighbors were Big Three engineers. Driving is a pretty dangerous thing to do, and I'm sympathetic to the challenge large automakers face when it comes to keeping the public safe. With large numbers, there will always be statistical anomolies to point at.

Manufacturers protect the public by using accelerated failure testing to "see" a 10-year lifetime in a matter of weeks or months. They do this by stressing the hell out of a part or system.

The experiments can be as simple as constantly running a set of engines for months at high RPM. Repeatedly stopping and starting engines for months. Perhaps applying heat, pressure, and stress to lug nuts or extreme levels of humidity to running electronic motors. During my statistics Master's degree, I had an excellent reliability analysis teacher, Luis Escobar who taught me how challenging and nuanced AFT can be, so I tend to give automakers benefit of the doubt. The lab will never completely mirror the real environment, so it's important for manufacturers to monitor the fleet once it's on the road.

That said, individual catastrophic failures are rare, and if the failures are unrelated then their joint probability will be infitesimal (e.g. 1/100 * 1/100 * 1/100 = 1 / 1 million). It's important that the public realize this and that automakers respond when a fleet of vehicles starts showing clusters of catastrophic failures. Given the rarity, I was surprised when the power steering stopped working while my wife was driving last week. It was the third catastrophic failure-type (lug nuts, throttle-body, and power steering) and none of these issues should ever happen in the lifetime of a vehicle.

The facts

  1. Ford reported the defect July 1st, 2015.
  2. The defect is caused by the TRW Generation 1 Electric Power Assisted Steering (EPAS) unit.
  3. Sometimes the result is sudden loss of power assist while driving.
  4. Ford identified 824,378 affected vehicles (2010 - 2012 Ford Fusions, Mercury Milans and Lincoln MKZs).
  5. Ford issued a recall for 393,623 (47%) vehicles (NHTSA Recall 15V-340, Ford No. 15S18).
  6. Ford excluded the 2010 model year from its recall.
  7. 12,551 complaints and/or warranty claims were made regarding the problem.
  8. Half of all complaints alleging accidents involved non-recalled vehicles.


The breakdown

I felt somewhat wronged and dissapointed. After all, my car has the same component but was excluded from the recall. The NHTSA identified 40 crashes related to the defect, half of which come from the excluded portion of vehicles. The NHTSA goes on to say that included vehicles have a projected 10-year failure rate of 8-14%, while excluded vehicles only 1%. Time to do some analysis!

The data

The NHTSA hosts defect investigations on their website. Ford submitted two databases as .zip files that include all 824,378 vehicles produced by partial VIN number and crucially production date. I created a repository to house my analysis, you can see the raw data here.

Next, I unpacked the databases (MS Access format, yuck!) and exported as portable SQLite and csv files.

## Source: local data frame [824,378 x 7]
##                  vin  make  model    MY                 DOB
##                (chr) (chr)  (chr) (dbl)               (chr)
## 1  3FAHP01A6ARXXXXXX  FORD FUSION  2010 2008-07-03 00:00:00
## 2  3FAHP01A8ARXXXXXX  FORD FUSION  2010 2008-07-03 00:00:00
## 3  3FADP0531ARXXXXXX  FORD FUSION  2010 2008-07-21 00:00:00
## 4  3FAHP01A5ARXXXXXX  FORD FUSION  2010 2008-07-21 00:00:00
## 5  3FAHP08A7ARXXXXXX  FORD FUSION  2010 2008-07-22 00:00:00
## 6  3FADP0536ARXXXXXX  FORD FUSION  2010 2008-07-23 00:00:00
## 7  3FAHP02GXARXXXXXX  FORD FUSION  2010 2008-07-23 00:00:00
## 8  3FAHP08A9ARXXXXXX  FORD FUSION  2010 2008-07-23 00:00:00
## 9  3FADP0534ARXXXXXX  FORD FUSION  2010 2008-07-24 00:00:00
## 10 3FADP0536ARXXXXXX  FORD FUSION  2010 2008-07-24 00:00:00
## ..               ...   ...    ...   ...                 ...
## Variables not shown: WSD (chr), SST (chr)

I organized, formatted and inspected both the production data and the failure data. No analysis or data cleaning is ever perfect. For example, a lack of production dates on the FMC360 failure data prevented me from including that dataset in my analysis. I'm curious how the NHTSA cracked that nut ...

Finally, I filtered out previously filed reports to ensure there weren't any duplicate reports in my analysis.

The analysis

After all the organizing and cleaning, once I was able to plot the data it became clear that 2011 and 2012 models have seen many many more failures than the 2010 model year. The 2011 and 2012 model years have about triple the failure rate by their third year of service. Oh great, I guess I'm just (un)lucky! On the otherhand, it's not as if the faiures are rare. They'll likely affect 1 in 100 2010 model year vehicles throughout a 10-year assumed lifetime.

The Fusion has seen its share of failures, the Mercury models have seen a much lower failure rate. Ironically the high-end Lincoln MKZ Hybrid has seen the worst power steering failure rates (hitting 1 in 50 by year four!).

Finally, looking at the failures by model and model year shows the full picture. The NHTSA predicts around 1 in 10 of the 2011/2012 model year vehicles will see power-steering failure (while driving) over a 10-year assumed lifetime.

Plotting failure report dates over time reveals the reason for the curvy nature of the 2011 and 2012 model year failure rates: failures cluster during the summer months. I'm curious about the reason this might happen, perhaps it's humidity / heat related?

Last weekend I built a little R API wrapper to the NHTSA API so that I (and others) could dig through the NHTSA recall and complain facilities.

In one final step, I wanted to sort the complaints in those facilities to see how power steering failures rank and what share of total complaints they represent.

Power steering failures (rank) and share of all complaints:

2010: (2) 10.5% of all complaints
2011: (1) 34.5% of all complaints
2012: (1) 52.6% of all complaints <--- wow, that's insane!

collate_failures <- function(year) {
  tbl_df(facility("complaints") %>%
           model_year(year) %>%
           vehicle_make("ford") %>%
           vehicle_model("fusion") %>%
           nhtsa_fetch()) %>%
    mutate(total = n()) %>%
    group_by(Component, total) %>%
    summarise(count = n()) %>% ungroup %>%
    mutate(share = count / total) %>%
    arrange(desc(share)) %>%
    mutate(share = percent(share)), 5) # show only the top 5

collate_failures(2009) # power steering not in the top 10
## Source: local data frame [5 x 4]
##                                       Component total count share
##                                           (chr) (int) (int) (chr)
## 1                         VEHICLE SPEED CONTROL   118     8 6.78%
## 2                                SERVICE BRAKES   118     7 5.93%
## 3                          STRUCTURE:BODY:DOOR    118     6 5.08%
## 5                                STRUCTURE:BODY   118     5 4.24%
collate_failures(2010) # steering failures @ 10.5% of all complaints
## Source: local data frame [5 x 4]
##               Component total count share
##                   (chr) (int) (int) (chr)
## 1           POWER TRAIN  2323   259 11.1%
## 2              STEERING  2323   245 10.5%
## 3 VEHICLE SPEED CONTROL  2323   219  9.4%
## 4                ENGINE  2323   124  5.3%
## 5  ENGINE | POWER TRAIN  2323   120  5.2%
collate_failures(2011) # steering failures @ 34.5% of all complaints
## Source: local data frame [5 x 4]
##               Component total count share
##                   (chr) (int) (int) (chr)
## 1              STEERING  1045   359 34.4%
## 2           POWER TRAIN  1045    71  6.8%
## 3 VEHICLE SPEED CONTROL  1045    65  6.2%
## 4                ENGINE  1045    53  5.1%
## 5  ENGINE | POWER TRAIN  1045    45  4.3%
collate_failures(2012) # steering failures @ 52.6% of all complaints
## Source: local data frame [5 x 4]
##               Component total count share
##                   (chr) (int) (int) (chr)
## 1              STEERING   572   301 52.6%
## 2           POWER TRAIN   572    37  6.5%
## 3                ENGINE   572    24  4.2%
## 4 VEHICLE SPEED CONTROL   572    21  3.7%
## 5              AIR BAGS   572    12  2.1%
collate_failures(2013) # steering failures @ 10.7% of all complaints
## Source: local data frame [5 x 4]
##                Component total count share
##                    (chr) (int) (int) (chr)
## 1               STEERING   197    21 10.7%
## 2         STRUCTURE:BODY   197    18  9.1%
## 3       UNKNOWN OR OTHER   197    11  5.6%
## 4               AIR BAGS   197    10  5.1%
## 5 FUEL/PROPULSION SYSTEM   197    10  5.1%
collate_failures(2014) # steering failures @ 13.5% of all complaints
## Source: local data frame [5 x 4]
##                Component total count share
##                    (chr) (int) (int) (chr)
## 1               STEERING    74    10 13.5%
## 2       VISIBILITY/WIPER    74    10 13.5%
## 3 FUEL/PROPULSION SYSTEM    74     5  6.8%
## 4      ELECTRICAL SYSTEM    74     4  5.4%
## 5               AIR BAGS    74     3  4.1%

The conclusion

After all of this, my lemon is still a lemon. It's currently at the dealer and tomorrow morning mechanics will be checking it out. When I spoke with Ford customer service over the phone they told me to have hope for a future recall as did the dealer. I'm not betting on it, but I do think it would be the right thing for Ford to do. Power steering failures are clearly numerous in the 2010 models despite being less than the 2011 and 2012 model years.

Vehicles Ford hasn't yet recalled account for half of all crashes alleged to be associated with this power steering failure defect. I can tell you first-hand that this is a very dangerous issue as are the lug nuts shearing off and the throttle body problem (diagnostic code P2111). If you have a 2010-2012 Ford Fusion, you should definitely make sure you have these defects fixed.

If you've experienced one of these issues yourself, I encourage you to file a complaint with the NHTSA. Complaints feed directly into statistical analyses that the NHTSA carries out to determine if a vehicle should be recalled.

As a final point, Ford's recall of 2011 and 2012 vehicles is woefully insufficient. The vehicle must display one of a subset of error codes to have the part replaced (so predictably next summer a bunch of these failures will happen, and the summer after). If your recalled vehicle doesn't display one of the certain codes, the remedy is a software update to improve the "auditory and visual warning systems." Yesh!

I'll conclude with some snippets of dealership reports taken from customers who experienced the power-steering defect first-hand:

Customer A

When the service power steering light comes on and you turn vehicle off then back on the power steering goes out totally. She has also had the throttle body replaced for the 2nd and is taking it to the dlr tomorrow for the recall. Feels as though the vehicle is unsafe to drive. Her husband hurt him self when the power steering went out the last time. Would like to talk to a supervisor. .... What are you seeking from Ford Motor Company? Just for the repairs to be taken care of. Would like for FMC to make the vehicle road worthy, not a road hazard. Wants a refund should a recall be issued for the multiple times she has had it repaired...6. What was the date of the accident? End of nov around thanksgiving...

Customer B

Experiencing problems with the power steering. The power sateering just cut off and was out for some time. The battery died this time and when he got a jump and the power steering is back on. Very concerned with the safety of this vehicle. Also has a throttle body concern, but he will be in contact with the dlr for diagnosis

Customer C


Customer D

CUSTOMER SAYS:..When driving the vehicle the power steering will fail. She said she has to pull over and shut the vehicle off and restart it to engage the power steering. She feels unsafe in the vehicle and is requesting repair assistance.....PER CUSTOMER DEALER SAYS:..There is no fix for this issue. They are unable to diagnose the vehicle as no code comes up.....

Customer E


Source: FMC360 database

Continue reading on