An Experimental Study on Validation Problems with

Existing HTML Webpages

Shan Chen Dan Hong Vincent Y. Shen
Department of Computer Science
Hong Kong University of Science and Technology
Clear Water Bay, Kowloon
Hong Kong, China
{chenshan, csdhong, shen}@cs.ust.hk

Abstract
In this paper we report the results of an experimental study on the
validation problem of existing HTML pages in the World Wide
Web. We have found that only about 5% of webpages are valid
according to the HTML standard. An invalid webpage may be
rendered differently by different browsers; it may not be machine-
processable; it might not be translated into other Web document
formats correctly using some of the translation engines available
on the Web; etc. Through sampling and analyzing the results, we
have identified the most common problems in existing webpages
that made them invalid. We hope our discovery can encourage
standard bodies, browser developers, authoring tool developers,
and webpage designers to work together so that the number of
valid webpages continues to grow and the Web can indeed reach
its full potential.
Categories and Subject Descriptors
H.5.2 [Information Interfaces and Presentation]: User
Interfaces - Standardization. I.7.5 [Document and Text
Processing]: Document Capture - Document analysis.
General Terms
Experimentation, Verification
Keywords
HTML Specification, HTML Validator, HTML tidy tools, W3C,
webpage characteristics, Web Sampling.
1. Introduction
Invented in the early 1990s, the World Wide Web is now one of
the primary media through which people publish and obtain
information. As one of the most popular applications on the
Internet, the number of webpages continues to grow rapidly. In
1999, there were 800 million webpages in the world [7], and
according to Internet Domain Survey of the Internet Systems
Consortium [6], there were already more than 233 million domains
by the end of J anuary 2004.
The importance of the Web attracted many researchers to study its
properties. People are interested in the average size of an HTML
document, the average number of images in a webpage, the
average size of a website, and so on [7] [8]. Our concern is about
the quality of the HTML documents in todays webpages in terms
of conformance to public standards.
The HTML is actually an extensible computer language used to
instruct the browser on how to render a webpage, which may
contain text, images, links, and multimedia contents. To gain a
competitive advantage in the browser battle each tool vendor
tried to introduce features that would distinguish its own products.
With these new features, webpage designers could easily create
exciting webpages, and browser users could enjoy these pages with
a brand-new experience. Unfortunately, features introduced by one
particular vendor might not be recognized by the browser from
another vendor. To create some order in a chaotic situation, the
World Wide Web Consortium (W3C) released the HTML (Hyper-
Text Markup Language) Recommendation in 1995 for resolving
interoperability concerns for browsers from different vendors [12].
Using the elements and attributes defined in the HTML
Recommendation, browsers parse the HTML document, analyze
the tags and render them (hopefully) as the webpage designer has
intended. This Recommendation has become the de facto
standard in the Web community.
The HTML documents that follow the standard or the valid
HTML files have a higher chance to be rendered consistently by
different browsers and across different platforms than those that do
not. An invalid HTML file may not be treated the same way. All
the major browser vendors claim to handle valid HTML files
correctly. It makes good sense for the developer of a new website
to start with the standard and follow it through later revisions.
In addition to possible different appearances when rendered,
invalid HTML files could be less accessible than valid ones. For
example, the standard defines a required attribute al t in
element I MG where al t provides some textual information
about the image. If the browser does not support images, the value
of the al t attribute is displayed instead. The textual information
can also be read out by a screen reader to assist the vision-
impaired user who cannot see the image clearly. An invalid
webpage missing the al t attribute may not be accessible by
certain users.
The Web has now become a large resource of information. More
and more people use the computer to extract relevant information
from the Web for their applications. An invalid HTML file often
causes difficulties for the application in understanding the intended
structure of the file. Information contained in such files may not be
available to these applications, defeating the purpose of pub-
licizing them on the Web. For example, the input interface of a

Copyright is held by the authors.
WWW 2005, May 10--14, 2005, Chiba, J apan.


This is the Pre-Published Version
Web application is usually presented as a FORM element in the
HTML file and the element TABLE is widely used to control the
layout of the page. Consequently webpage designers often use
FORM and TABLE together to provide interactivity as well as
to control the layout. However, if the webpage designer does not
understand the HTML language well, it is easy to commit errors
while using FORM with TABLE. Such errors may prevent the
interactive page to be machine-processable.
To assist webpage designers, several organizations have developed
validators to check if a webpage conforms to the HTML
standard. These validators can be used to analyze a webpage (for
example, the W3C markup services [14] and the WDG validator
[16]), or to analyze a whole website (for example, the CSE HTML
Validator [1]). There are also a number of HTML tidy tools
which can be used to tidy up an HTML document by correcting
most of the errors in it [10]. However these tools are often not
powerful enough to correct all errors; sometimes the appearance of
a webpage is altered after being tidied up. Proofreading the
resulting document after processing by a tidy tool could be a
challenge to the webpage designers.
We wish to find out the proportion of current webpages that are
valid HTML documents, and the most common problems that
cause a document to become invalid. This issue was of interest to
the Web Design Group; they identified six common problems
[15] by inviting comments from other validator users. However,
there was no indication how serious each problem was, and there
was also no report on the methodology used by various individuals
in identifying these problems. It is difficult to draw quantitative
conclusions from these experiential results. To really understand
how serious the validity problem on the Web is today, and to study
the statistics of validation problems, we performed an
experimental study by sampling the existing webpages, using a
validator to check them, and analyzing the results. Our study
shows that only about 5% of webpages are valid. We also
identified the major validation problems in existing HTML
webpages. We hope our findings will alert the webpage designers
of the validity problem, and will help them to identify the most
common errors which they should try to avoid. We hope our
findings will also alert the Web application developers so that they
could handle the common errors properly in order to provide better
service. We also hope our findings are useful to the standard
bodies and the authoring tool vendors.
The rest of the paper is organized as follows. Section 2 presents
how we sampled the World Wide Web and describes the set up of
our experiment. Section 3 presents experimental results, showing
only about 5% of the webpages are valid. It also identifies the most
common problems causing an HTML file to fail the HTML
validator. Section 4 analyzes the reasons which caused these
problems and provides some possible solutions. We present our
conclusions in Section 5.
2. Experiment Set Up
Our main purpose is to study the common problems in current
webpages. If we can test all the existing webpages, then the results
will be accurate and convincing. Unfortunately this is not feasible
since there are too many pages on the Web. If we could test
random samples of this huge population, the test results would
reflect the state of the existing webpages.
The unbiased selection of a random subset of the Web has been an
open issue. In 1996 Bray [4] used a self-defined start set of
about 40,000 webpages, crawled the Web to measure it, and
obtained some statistics on the size of the Web, its connectivity,
richness, and the distribution of URL links. Bharat and Broder [3]
suggested a methodology for finding a page at random from a
search engine index in 1998. Their approach was based on queries
to the search engine using random words. In 1999 Lawrence and
Giles [7] used randomly-generated IP addresses to estimate the
information available on the Web, such as the characteristics of the
hosts and pages. In 2000 Henzinger, Heydon, Mitzenmacher, and
Najork [5] suggested performing a random walk of the Web using
multiple search engines to extract representative samples of
webpages.
In this paper, we adopted the methodology of using randomly-
generated IP addresses as samples due to its simplicity. We further
compared our test results using this Random IP List with the result
using the Most Popular 10000 Websites List made available to us
by Alexa.com as additional supporting evidence. Note that both
lists enabled us to analyze the properties of only the home pages of
the selected websites. Although problems with the homepage may
be indicative of the problems of other webpages accessible in that
website, we created yet another URL List using search engines to
confirm our findings. We repeated the test using the Second
Random IP List in two months to see if there is any change of
results over time. The details of the lists are described below.
2.1 Sample Lists
2.1.1 Random IP List
We developed a program to randomly generate an IPv4 address.
By setting up an HTTP connection to this IP address at port 80,
which is the default port for HTTP connections, and listening to
the response, we would know whether it was a Web server at this
IP address. If it was indeed a Web server, we added it to the list.
We encountered some difficulties using this approach. Since most
IP addresses were addresses for online computers which were not
Web servers, most of our probes would time out. After a short
while our service provider received complaints about our activities
since our probes were considered hackers probes by some
organizations. We had to stop the collection process after
obtaining a list of only about 1,100 IP addresses.
2.1.2 Most Popular 10000 Websites List
There are several URL lists available on the Web, such as the 500
most popular websites from Alexa [2]. The validity of such
websites is important since, if ranked correctly by the list
providers, they are frequently accessed by individuals or
computers. Their properties have higher impact on rendering,
accessibility, and interoperability issues than a webpage selected
randomly. We appreciate the support of Alexa.com who, after
understanding the goals of our research project, provided us with
the Most Popular 10000 Websites List. The results obtained using
this list supported our findings using the Random IP List.
2.1.3 URL List
Both the Random IP List and the Most Popular 10000 Websites
List allowed us to check the home pages of these websites only.
Since homepages constitute only a small portion of the whole
Web, we wish to study the properties of other webpages to see if
they are consistent with the properties of homepages.
Since most people access webpages using search engines, studying
the properties of a sample of webpages indexed by some search
engine will provide additional supporting data to our findings.
Although the indexing method is proprietary information by the
search engine owner, we could use a method similar to that used in
[3] to obtain a sample. Instead of sending random words to a
search engine, which may limit the responses to webpages with
English contents only, we used a random string to match URLs of
webpages indexed by that search engine.
According to RFC1738 [9], a URL may contain English letters,
digits, and certain special characters (i.e., any of ; / ?: @=&$-
_. +! * ( ) , %). We generated a string of up to eight such
characters randomly and sent it to a search engine which supported
URL search, and included the results in our URL list. If a
particular search request yielded more than ten hits, we included
the first ten in our list. The search engines that we used are
Google, Yahoo and Teoma. We assume they all independently
developed their own indices. We realized that this list was
certainly not a random sample of webpages but used that anyway
to confirm our findings. Note that this list of 31,540 URLs did not
contain our own biases.
2.1.4 Second Random IP List
We also repeated the random IP address generation methodology
two months later and got the Second Random IP List over a period
of one month. We used two different machines with forty threads
each and managed to bypass the restrictions imposed by our
service provider earlier. This list has 1,700 URL addresses.
2.2 The Experiment
A Validation Test program was developed in J ava. Two major
tools were used in this program: the W3C Markup Validator [14]
and the J Tidy parser [10].
The W3C Markup Validation Service is a free service that checks
documents like HTML and XHTML for conformance to W3C
Recommendations and other standards [14]. The validator uses a
CGI script to fetch the given URL, parses it, and post-processes
the resulting error list for easier reading. It returns a webpage
which contains the error messages.
The J Tidy parser is a J ava port of HTML Tidy, an HTML syntax
checker and pretty-printer. J Tidy can be used as a tool to clean up
malformed or invalid HTML documents. The version of J Tidy we
used is 04aug2000r7-dev.
Figure 1 shows the workflow of our experiment.
Internet
W3C Validator
Web Server
Lists
URL
URL
Original
HTML files
Parser
URL
Validation
result page
Information collection
about the original
webpage
Information collection
about validation errors
Internet
W3C Validator
Web Server
Lists
URL URL
URL
Original
HTML files
Parser
URL
Validation
result page
Information collection
about the original
webpage
Information collection
about validation errors

Figure 1 Workflow of our test tool
Our program sequentially picks up a URL in a sample list. The
original webpage returned by the Web server is parsed by the
J Tidy parser to get certain information about the webpage tested:
the I MG counter, the FORM counter, the W3C validation icon
counter, etc. Then the URL is sent to the Validator, and the
returned result page is parsed to record the error information. We
stored the results in a file for further analysis.
After we got the error logs from testing the sample lists, we were
able to draw an initial conclusion of the common validation errors.
3. Experimental Results
With the Web sample lists obtained in Section 2, we used the W3C
validator
1
to check the errors.
3.1 Analysis of results from the Random IP
List
3.1.1 Error Definitions
From the webpages we tested, our program was able to catch about
98% of the total errors reported by the W3C Validator. Figure 2
shows the error distribution in the Random IP List. Note that our
list includes the six common problems [15] identified by WDG.
A more detailed analysis shows that some of these six problems
are not affecting a significant number of existing webpages.
We explain some of the most common errors below. All the
figures shown in this section are based on the data from the
Random IP List.
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
Common Errors
Missing DOCTYPE
Specification
Missing Required
Attributes
Invalid Attributes
Missing Encoding
Specification
Wrong Placement of
Element
Undefined Elements
Missing Quotation Marks
Ampersands (&) in URL
Wrong Value of Attribute

Figure 2 Most common errors
3.1.1.1 Missing DOCTYPE Specification
The DOCTYPE statement is a machine-readable statement in the
HTML document which specifies the structure, elements, and
attributes used in that document [11]. The W3C QA Activity
maintains a List of Valid Doctypes [13] to choose from. A valid
DOCTYPE is given below:
<! DOCTYPE HTML PUBLI C
" - / / W3C/ / DTD HTML 4. 01 Tr ansi t i onal / / EN"
" ht t p: / / www. w3. or g/ TR/ ht ml 4/ l oose. dt d" >

1
Since May 2004, W3C updated their validator several times. The
validators we used are V0.6.5, V0.6.6 and V0.6.7. However,
these validators did not cause much difference in the results.
We found that more than 50% of the webpages omitted this
important specification. Note also that the formal public
identifier the quoted string that appears after the PUBLI C
keywordis case-sensitive. An error would occur if the following
is given instead:
<! DOCTYPE HTML PUBLI C
"-//w3c//dtd html 4.0 transitional//en">
Since so many webpages did not have the DOCTYPE
specification, the validator automatically chose the most tolerant
specification (HTML 4.01 Transitional) before proceeding to parse
and analyze the HTML document.
3.1.1.2 Missing Required Attributes
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
35.0%
40.0%
45.0%
alt type action content
Missing Required Attributes

Figure 3 Missing Required Attributes errors
There are several attributes required in specific elements: for
example, the attribute al t of I MG and the attribute t ype
of SCRI PT. Since omitting these attributes causes no rendering
difficulties for browsers, webpage designers usually neglect to
include them. In Figure 3, we found that nearly 40% of webpages
omitted the attribute al t while 30% of webpages omitted
t ype. Even when it provides the al t , the value of al t is
often an empty string. We also noticed that nearly 60% of the
element SCRI PT omitted t ype. These two kinds of attribute
errors made up 99% of the Missing Required Attributes errors.
About 60% of all the webpages have this kind of errors.
A test on the quality of the text for the al t attribute is provided
by http://www.hisoftware.com/accmonitorsitetest/. By using this
validator, we find that only 38% webpages in the Random IP List
passed this test, 17% webpages passed the test with some
warnings, and the remaining 44% failed the test.
3.1.1.3 Invalid Attributes
In addition to omitting some required attributes, webpage
designers sometimes include attributes that are not defined in the
HTML standard, or defined but not for the elements they are using.
Figure 4 shows the distribution of these invalid attributes. These
errors can be divided into two types: those not defined in the
HTML standard and those defined but not for the elements they
are used.
Attributes such as t opmar gi n, l ef t mar gi n, mar gi n-
hei ght , mar gi nwi dt h and bor der col or are not
defined in the HTML standard. Browsers may render the webpage
differently when these attributes are used in a particular element.
Attributes such as hei ght , wi dt h, bor der , al i gn,
backgr ound, and val i gn are valid attributes in HTML.
However, when these attributes are used in some elements, it may
cause validation errors. For example, backgr ound is a valid
attributes for BODY but <TABLE backgr ound=
" t est . j pg" >is invalid since backgr ound is not a defined
attribute for TABLE.

0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
h
e
i
g
h
t
M
a
r
g
in
P
r
o
b
l
e
m
b
o
r
d
e
r
b
a
c
k
g
r
o
u
n
d
w
i
d
t
h
ty
p
e
c
o
lo
r
b
o
d
e
r
-
c
o
lo
r
n
a
m
e
s
r
c
q
u
a
l
i
ty
p
lu
g
i
n
s
p
a
g
e
b
g
c
o
lo
r
v
a
li
g
n
a
l
ig
n
Invalid Attributes

Figure 4 Invalid Attributes errors
Some of these problems are caused by certain webpage authoring
tools. For example, Microsoft FrontPage, which is a widely used
authoring tool, adds attributes like t opmar gi n, l ef t -
mar gi n, mar gi nhei ght , and mar gi nwi dt h to the
BODY element automatically when a user tries to define the page
properties. This is called the Margin Problem by us. Such
special assistance to the webpage designers makes the document
invalid.
We found that theInvalid Attributes errors occur in about 60% of
the webpages.
3.1.1.4 Missing Encoding Specification
To improve interoperability, each HTML document needs to
specify a character set for the document, which is called character
encoding in the HTML standard. If an HTML document does not
provide the encoding information (for example, a traditional
Chinese character webpage), the browsers may not render the
characters correctly on the webpages. The users must manually
choose the encoding while browsing, which is inconvenient.
In order to validate an HTML file without encoding specification,
the Validator automatically chooses UTF-8 encoding that contains
most known characters. However, some characters in ISO-8859-
1(Western Europe), which is another commonly used encoding
specification, are illegal or incompatible with UTF-8. If the actual
encoding is ISO-8859-1 and the encoding specification is not
provided in the HTML document, the Validator may fail to
validate the file. Therefore, in our experiment we tried to let the
Validator use ISO-8859-1 as the encoding to validate if the earlier
attempt to use UTF-8 failed.
3.1.1.5 Wrong Placement of Element
Not only do the elements and attributes undefined in the HTML
standard cause validation errors, defined elements may also cause
errors if placed incorrectly. For example, it is wrong to place
META or LI NK element in the BODY section since they
should be placed inside the HEAD section of an HTML
document. Furthermore, some elements need to be placed in a
containing element. One possible cause for this kind of errors is
that a block-level element (such as P or TABLE) is placed
inside an inline element (such as A, SPAN, or FONT). A
Wrong Placement of Element error may also occur if a previous
element is not closed. About 35% of webpages have this kind of
problems. The details are shown in Figure 5.
0.0%
1.0%
2.0%
3.0%
4.0%
5.0%
6.0%
7.0%
P
T
R
T
D
H
R
H
1
-
H
6
T
A
B
L
E
L
I
N
K
S
C
R
I
P
T
D
I
V
M
E
T
A
S
T
Y
L
E
B
R A
Wrong Placement of Element

Figure 5 Wrong Placement of Element errors
One of the Wrong Placement of Element errors merits special
attention: Form with Table. As defined in the standard, FORM
should be specified in a container, like TD, instead of being
specified as an element within TABLE. Since TABLE is often
used to control the page layout, and FORM is often used for Web
applications requiring user input, these two elements often are used
together. At least 15% of FORM elements were in error when
FORM was specified within TABLE, rather than in the TD
within that TABLE. Unfortunately, the webpage designer might
not notice the error since most browsers would render the webpage
in the same look-and-feel as the designer had intended. We also
noticed that nearly 10% of the FORM elements did not have the
required act i on attribute; they were probably put there as a
place holder but caused a validation error.
3.1.1.6 Undefined Elements
Before the HTML standard was established some browser vendors
defined elements that would be useful in creating exciting
webpages. One such element is EMBED, which could be used to
insert music or video in the HTML document. Although EMBED
has been replaced by OBJ ECT in the standard, many legacy
documents still use it and most browsers still support it.
Nevertheless a webpage containing EMBED is considered invalid
by the Validator. Other frequently-used elements that are
undefined in the standard include NOBR and SPACER.
3.1.1.7 Missing Quotation Marks
The HTML standard requires quotation marks around literal values
of attributes. For example, a white background color should be
specified as bgcol or =" #FFFFFF" . Although most browsers
can render the color correctly with or without quotation marks,
they may render differently in other situations such as
si ze=" +1" and si ze=+1. Therefore missing quotation
marks could cause a webpage to become invalid. We were
surprised to note that when Microsoft Internet Explorer is used to
save an HTML document using the function Save As Web Page,
complete, it removes most of the quotation marks around literal
values. This browser may be the culprit of many invalid webpages.
3.1.1.8 Ampersands (&) in URL
The ampersand (&) is often used in URLs. For example,
ht t p: / / www. f oo. com/ user =t est &pswd=t est
But if this URL is used as a hyperlink in an HTML document, it
would generate an error since & is a reserved symbol in HTML
that is the beginning of an entity. It should be changed to
&amps; in the hyperlink like the following:
<a hr ef =
" ht t p: / / www. f oo. com/ user =t est &amps; pswd=t est "
></ a>
3.1.1.9 Wrong Value of Attributes
0.0%
1.0%
2.0%
3.0%
4.0%
5.0%
6.0%
align name valign nowrap
Wrong Value of Attributes

Figure 6 Wrong Value of Attributes errors
In the HTML standard some attributes have a specific range of
values. For example, the value of al i gn only has five choices:
bot t om, mi ddl e, t op, l ef t , and r i ght . But
webpage designers often specify its value to be cent er ,
making the webpage invalid. The attributes val i gn and
name also have a similar problem. Figure 6 shows the
distribution of such errors.
3.1.2 Top Ten Problems for HTML files in Random
IP List to fail the HTML validator
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
Top ten with RandomIP List
MissingDOCTYPE
Specification
MissingAttribute "alt"
MissingEncoding
Specification
MissingAttribute "type"
InvalidAttributes of "height"
and"width"
MarginProblemin"BODY"
MissingQuotation Marks
InvalidAttribute "border"
InvalidAttribute "background"
'&' InURL

Figure 7 Top ten problems in the Random IP List
From the analysis above, we are able to list the top ten problems
that caused the HTML documents to fail the validator:
Missing DOCTYPE Specification
Missing Attribute alt within Element IMG
Missing Encoding Specification
Missing Attribute type with Element SCRIPT or
STYLE
Invalid Attributes height and width
Margin Problem: invalid extra attributes topmargin,
leftmargin, marginheight, and margin-
width in element BODY
Missing Quotation for Attribute Values
Invalid Attributes border
Invalid Attributes background
Ampersands in URLs
3.2 Problems for the Most Popular 10000
Websites List
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
Top ten with Most Popular 10000 Websites List
Missing DOCTYPE
Specification
Missing Attribute "alt"
Missing Attribute "type"
Invalid Attributes "height"and
"width"
'&' InURL
MarginProblemin"BODY"
Missing QuotationMarks
Invalid Attribute "background"
Invalid Attribute "border"
Formwith Table

Figure 8 Top ten errors of the Most Popular 10000 Websites
List
We wish to confirm our findings using the Most Popular 10000
Websites list from Alexa.com. The results are shown in Figure 8.
Note that the top 10 problems are very similar to that from the
Random IP List, except that the ordering has been changed
somewhat. The exception is that Missing Encoding Specification
has dropped out of the top ten. The Form with Table problem,
which was number 33 from the Random IP List, enters the top 10.
We believe the webpage designers of the most popular websites
are committed to make their contents readable by people around
the world. Therefore they included the encoding specification so
that users need not select the encoding manually.
Since many of these popular websites interact with their users,
such as getting information from them, they use FORM. They
may also use TABLE to control the layout. Since most popular
browsers support invalid usage of FORM and TABLE, such
errors might not have been noticed by the webpage designers of
these popular websites.
Another interesting thing is that the error rate in the Most Popular
10000 Websites list is relative higher than the error rate in the
Random IP List. That might be the result of frequent updates; any
mistake due to an update is likely to stay unless it causes rendering
difficulties.
3.3 Problems for the URL List
In previous tests using the Random IP List and the Most Popular
10000 Websites List, we only tested the homepages. Do all the
pages have problems similar to that shown in Figure 7 and Figure
8? We used the URL List, which contained both the homepages
and linked pages. The results are shown in Figure 9.
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
Top ten with URL List
MissingDOCTYPE
Specification
MissingAttribute "alt"
MissingAttribute "type"
Invalid Attributes of "height"
and"width"
MissingEncoding
Specification
'&' In URL
MissingQuotation Marks
Margin Problemin "BODY"
Invalid attibute "border"
Invalid attibute "background"

Figure 9 Top ten errors in the URL List
From Figure 9, we can find that the distribution of the top ten
errors is the same as that in the Random IP List. The only two
differences between Figure 7 and Figure 9 are the ranking of errors
and the error rates.
The Form with Table problem is not one of the top problems
compared with Figure 8. Linked pages usually provide information
about some particular topics. The need for user interactions in
these inner pages is much less compared to the home page of a
website. That might have reduced the occurrences of such errors.
3.4 Further confirmation with a new Random
IP List
From the analysis, we find that the top ten errors in HTML
validation problem are almost the same, no matter what collection
of webpages was used. We used the same method to generate the
Second Random IP List and did the validation test three months
later. The results are presented in Figure 10.
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
Top ten with Second RandomIP List
Missing DOCTYPE
Specification
Missing Attribute "alt"
Missing Encoding
Specification
Missing Attribute "type"
InvalidAttribute "height" and
"width"
Margin Problemin "BODY"
Missing Quotation Marks
Invalidattibute "border"
Invalidattibute "background"
'&' In URL

Figure 10 Top ten errors of the Second Random IP List
Comparing Figure 7 and Figure 10, we can see that the top
problems in the Second Random IP List are the same as the top
problems in the Random IP list and the only difference between
each other is the minor difference in error rates. After getting this
data, we are more confident about the top problems that caused the
HTML files to become invalid. We noticed that the top ten
problems were stable although the error rate could be different.
There were also minor variations in ranking.
4. Discussion
4.1 Causes of the validation problems
From the experimental results we believe there are three main
reasons for an HTML file to be invalid: tolerant browsers,
authoring tools that introduce non-standard features, and the
culture of website development.
4.1.1 Tolerant Browsers
The Web was already a vibrant media for information exchange
when W3C released the HTML recommendation in 1995. Some of
the existing websites had used certain features that were not later
included in the standard. Although the browser developers
considered the standard and would handle valid webpages
correctly, as commercial product vendors they needed to satisfy
the needs of their customers; i.e., to render the webpage nicely if at
all possible. If a webpage designer, who may not be familiar with
the standard, uses a feature that deviates from the standard, there is
a good chance that the webpage will still be rendered nicely when
tested with some popular tolerant browsers. Webpage designers
are only concerned with whether browser users can get the
expected rendering effect. Once this is confirmed on the popular
browsers, webpage designers are reluctant to improve the code
further. Such mistakes will then become part of the Web legacy.
This would not have been a problem if the Web is mainly used in
human-computer interactions. Tolerant browsers contribute to the
problem of computer-computer interactions on the Web, which is
keeping the Web from reaching its full potential.
4.1.2 Zealous Authoring Tools
A great contribution of the World Wide Web is that people who
are not trained in computer programming can create exciting
websites which are appreciated by others. Web authoring tools
(such as Macromedia Dreamweaver, Microsoft FrontPage,
Netscape Composer, etc.) are of great help to those who are not
computer professionals. The authoring tools assist the webpage
designer to layout the website as easily as using a word processor
or editor. Unfortunately some of them, possibly for reasons of
aesthetics, introduce features that deviate from the HTML
Recommendation. We mentioned the example of Microsoft
FrontPage which adds invalid attributes such as mar gi n-
hei ght , mar gi nwi dt h, l ef t mar gi n, etc. to the
element BODY automatically when the page property is defined.
These introductions, coupled by tolerant browsers, contribute to
the problem of computer-computer interactions on the Web. There
seems to be a trend that the situation may be getting worse; for
example, Microsoft is introducing DHTML, which is an extension
of HTML that includes features on multimedia, database access,
and a new object model. It will cause chaos on the Web if such
features are not accepted as standard in a future version of the
HTML Recommendation.
4.1.3 Culture of Website Development
Since many website developers are not computer professionals, a
common practice in website development is to copy the look and
feel of interesting websites and then change their contents.
Mistakes in legacy websites are copied in this process. They are
not identified and fixed since the tolerant browsers do not reveal
them as problems. These new but invalid websites join the legacy
of the Web and further propagate the problems.
4.2 Possible solutions
Although a number of the problems we identified, such as the
Margin Problem, are cosmetic in nature, there are problems that
delay the Web in reaching its full potential. For example, the
missing al t attribute will make the website less accessible.
Some problems, such as Wrong Placement of Element, may cause
rendering errors or even failure in parsing the HTML document.
This will affect the processing of the document by computers.
The Web is a most popular application since it brings the benefit
of computers to everyone who has access to the Internet. We
cannot expect the user community to solve the problem, since
many webpage designers are not familiar with the HTML standard
and, even if they do, they may not know how to fix the mistakes.
Due to the huge legacy of existing webpages we cannot expect the
browser vendors to reveal problems since if some browser cannot
handle some popular non-standard features, people would not use
them.
We hope our findings are useful to the standard bodies and
authoring tool vendors. They should review the existing practices
and reach consensus on which features should become part of the
standard, and which features should no longer be supported by
authoring tools. The authoring tools could do better by including
the feature of some Tidy tools before the work is saved. At least
they should not provide hints that would suggest the use of invalid
features while the page is being created. If all authoring tools can
support that the validity check before saving a document, not only
new webpages will be valid but also some existing webpages,
when updated or copied using the authoring tool, will be valid
also. It is only through the better cooperation of standard bodies
and authoring tool vendors can the Web reach its full potential.
We also noticed that an interactive process between the tidy tool
and webpage designers would be better than the existing approach.
For example, the tidy tool inserts al t =" " when an image
misses the attribute al t . This will not help users to learn more
about the image if it could not be displayed or viewed by vision-
impaired people. If the tidy tool can remind the webpage designer
to write a brief introduction about the image, this problem can be
fixed.
5. Conclusion
The validation problems in existing webpages have drawn more
and more attention because of the increasing trend of Web
communications moving from a human-to-computer process to a
computer-to-computer process. In this paper, we conducted an
experiment that identified the major validation problems in current
webpages. We use the Random IP List as a basic test sample and
identify the top ten problems which cause the webpages to fail the
validator. We further use the Most Popular 10000 Websites List,
the URL list and the Second Random IP List to confirm our
findings. It is disappointing to find that only about 5% webpages
in our sample lists are valid according to our test results. Some
problems, such as the missing al t and the FORMwith TABLE
problems not only affect Web accessibility but also affect the
move to computer-to-computer communications on the Web. We
believe our findings are useful to the standard bodies and authoring
tool vendors, as well as to the Web application developers and
webpage designers. Due to the diversity of validation problems, it
is not going to be easy to fix all the errors in existing webpages.
But we can improve the situation by removing the causes of the
most common problems as suggested in our paper. Although the
webpages may still be invalid after fixing the most common errors,
we believe that through a deliberate effort and with the
collaboration of webpage designers, standard bodies, and vendors,
the validation problems could be gradually reduced, and the Web
can eventually reach its full potential.
6. Acknowledgement
This project was supported in part by the Sino Software Research
Institute grant No. SSRI 01/02.EG14, World Wide Web Consor-
tium Office for China.
7. References
[1] AI Internet Solutions, CSE HTML Validator.
http://www.htmlvalidator.com/.
[2] Alexa.com, 500 Most Popular Web Sites.
http://www.alexa.com/site/ds/top_500?p=DestTrLearn_W_g_
40_T1, May, 2004.
[3] Bharat, K., and Broder, A., A technique for measuring the
relative size and overlap of public web search engines.
Proceedings of the 7
th
World Wide Web Conference
(Brisbane Australia, May 1998), 379-388.
[4] Bray, T., Measuring the web. Proceedings of the 5
th
World
Wide Web Conference (Paris France, May 1996).
[5] Henzinger, M., Heydon, A., Mitzenmacher, M., and Najork,
M., On near uniform url sampling. Proceedings of the 9
th

World Wide Web Conference (Amsterdam The Netherlands,
May 2000), 295-308.
[6] Internet Systems Consortium. ISC Internet Domain Survey.
http://www.isc.org/ops/ds/, Dec 2003.
[7] Lawrence, Steve, and Giles, C. Lee, Accessibility of
information on the web. Nature 400 (J uly 1999), 107-109.
[8] Lee, D.C., and Midkiff, S.F. A sample statistical
characterization of the world-wide web. Proceedings of
Southeastcon 97. Engineering new New Century (April
1997), 174-178.
[9] Lee, T. Berners, Masinter, L., and McCahill, M., RFC1738.
http://www.ietf.org/rfc/rfc1738.txt, December 1994.
[10] SourceForge, J Tidy Project. http://jtidy.sourceforge.net/, May
2004.
[11] W3C, Dont forget to add a doctype.
http://www.w3.org/QA/Tips/Doctype, August 2002.
[12] W3C, Hypertext makeup language activity statement.
http://www.w3.org/MarkUp/Activity, J anuary 2003.
[13] W3C, List of Valid Doctypes.
http://www.w3.org/QA/2002/04/valid-dtd-list.html, Dec
2003.
[14] W3C, W3C Makeup Validation Service.
http://validator.w3.org/, May 2004.
[15] Web Design Group, Common HTML Validation Problem.
http://htmlhelp.com/tools/validator/problems.html, May 2004.
[16] Web Design Group, WDG validator.
http://www.htmlhelp.com/tools/validator/